View
1
Download
0
Category
Preview:
Citation preview
Virginia Commonwealth University Virginia Commonwealth University
VCU Scholars Compass VCU Scholars Compass
Theses and Dissertations Graduate School
2014
QUANTUM EFFICIENCY ENHANCEMENT FOR GAN BASED LIGHT-QUANTUM EFFICIENCY ENHANCEMENT FOR GAN BASED LIGHT-
EMITTING DIODES AND VERTICAL CAVITY SURFACE-EMITTING EMITTING DIODES AND VERTICAL CAVITY SURFACE-EMITTING
LASERS LASERS
Fan Zhang
Follow this and additional works at: https://scholarscompass.vcu.edu/etd
Part of the Electrical and Electronics Commons, and the Electronic Devices and Semiconductor
Manufacturing Commons
© The Author
Downloaded from Downloaded from https://scholarscompass.vcu.edu/etd/3655
This Dissertation is brought to you for free and open access by the Graduate School at VCU Scholars Compass. It has been accepted for inclusion in Theses and Dissertations by an authorized administrator of VCU Scholars Compass. For more information, please contact libcompass@vcu.edu.
VIRGINIA COMMONWEALTH UNIVERSITY
QUANTUM EFFICIENCY ENHANCEMENT
FOR GAN BASED LIGHT-EMITTING DIODES AND VERTICAL CAVITY SURFACE-EMITTING
LASERS
A research dissertation submitted in partial fulfillment of the requirements for the degree
of Doctor of Philosophy in Electrical and Computer Engineering at Virginia Commonwealth University
By
Fan Zhang
Director: MORKOÇ, HADIS FOUNDER PROFESSOR OF ELECTRICAL AND COMPUTER ENGINEERING
Committee in charge
PROF. MORKOÇ, HADIS
PROF. ÖZGÜR, ÜMIT
PROF. RESHCHIKOV, MICHAEL
PROF. YE, DEXIAN
DR. AVRUTIN, VITALIY
Dec 2014
ii
Acknowledgement
This dissertation would not have been possible unless the help and support of many
people who are gratefully acknowledged here.
I want to express my great gratitude to my dedicated advisor Professor Hadis Morkoҫ
for inspiration and mentoring. It has been a privilege to have been educated by him. He
has provided precious and deep ideas, comments and critiques with his profound
knowledge in nitride semiconductors and rich research experiences. I am very much
thankful to his help during my PhD study.
I would also like to acknowledge Prof. Ümit Özgür, Dr. Vitaliy Avrutin and Dr. Natalia
Izyumskaya for their help and guidance during my research. It was a great pleasure and
memory to have worked with them.
I would like to thank other committee members: Prof. Dexian Ye and Prof. Michael
Reshchikov. It is a great honor to have them on my committee. Thank you for
participating in my PhD journey and your valuable comments and suggestions.
Appreciation is also due to all research collaborators in my lab and other universities.
Thanks my lab fellows Xianfeng Ni, Mo Wu, Xing Li, Congyong Zhu, Serdal Okur,
Huiyong Liu, Shopan Hafiz, Morteza Monavarian, Mykyta Toporkov, Nuri Can, Barkat
Ullah, Mahbub Alam and Saikat Das for their help on my experiments. Thanks Daniel
Rosales and Professor Bernard Gil from University of Montpellier and Dr. Sebastian
Metzner, Professor Frank Bertram and Professor Juergen Christen from University of
Magdeburg. Our monthly videoconferences helped me to increase my understanding on
optical properties of GaN-based optoelectronic devices.
iii
Finally, I dedicate my thesis to my parents and my parents-in-law for their unconditional
love and support. Their high expectations have been the real driving forces behind me for
so many years. Last but not least, I would thank my wife Jingfei Tian, thank you for
giving me the endless supports and love in every possible way.
iv
Table of contents
Acknowledgement .............................................................................................................. ii Table of contents ................................................................................................................ iv List of Tables ..................................................................................................................... vi List of Figures ................................................................................................................... vii Abstract .............................................................................................................................. xi Chapter 1 Introduction ................................................................................................... 1
1.1 Motivation ............................................................................................................ 1 1.2 Development in GaN-based light-emitting diodes ............................................... 1 1.3 Development in GaN-based vertical cavity surface-emitting lasers .................... 4 1.4 Scope of research ................................................................................................. 7
Chapter 2 Efficiency droop investigations .................................................................... 9 2.1 Introduction to the efficiency droop ..................................................................... 9 2.2 Hot electron model ............................................................................................. 11 2.3 Stair-case injector (SEI) designs ........................................................................ 13 2.4 Graded Electron Injector (GEI) .......................................................................... 18 2.5 Electron leakage from active region ................................................................... 20
2.5.1 Diffusion length measurement in p-GaN .................................................... 20 2.5.2 Electron escape from active region ............................................................. 25
2.6 Carrier overflow versus Auger recombination ................................................... 26 Chapter 3 Quantum efficiency enhancement in GaN-based LEDs ............................. 33
3.1 Effects of quantum barrier: height and thickness ............................................... 33 3.1.1 Motivation ................................................................................................... 33 3.1.2 Experimental procedures ............................................................................ 33 3.1.3 Results and discussion ................................................................................ 35
3.2 Quantum efficiency for multi-DH LEDs ........................................................... 37 3.2.1 Motivation ................................................................................................... 37 3.2.2 Experimental procedures ............................................................................ 39 3.2.3 Results and discussions ............................................................................... 41
3.3 Optimization of stair-case electron injector ....................................................... 47 3.3.1 Motivation ................................................................................................... 47 3.3.2 Results and discussions ............................................................................... 48
3.4 Delta p-doped barrier in LED active regions ..................................................... 55 3.4.1 Motivation ................................................................................................... 55 3.4.2 Experimental procedures ............................................................................ 56 3.4.3 Numerical Simulation of delta p-doped barrier LEDs ................................ 58 3.4.4 Results and discussion ................................................................................ 61
3.5 Investigation of GEI LEDs ................................................................................. 63 3.6 Injection-dependent radiative recombination coefficient (B) of single and multi active layer DH LEDs ................................................................................................... 65 3.7 Investigation of layer quality on semi-polar GaN .............................................. 76
3.7.1 Motivation ................................................................................................... 76 3.7.2 Experimental procedure .............................................................................. 76 3.7.3 Results and discussion ................................................................................ 77
v
Chapter 4 Quantum efficiency enhancement in GaN-based VCSELs ........................ 81 4.1 Hybrid VCSELs ................................................................................................. 81
4.1.1 Brief introduction to hybrid VCSELs ......................................................... 81 4.1.2 Experimental procedures ............................................................................ 82 4.1.3 Results and discussions ............................................................................... 84
4.2 GaN-based VCSELs with both dielectric DBRs ................................................ 87 4.2.1 Motivation ................................................................................................... 87 4.2.2 Experimental procedures ............................................................................ 88 4.2.3 Optical characterization for VCSEL ......................................................... 100 4.2.4 Fabrication and electrical characterization for VCSEL ............................ 102
Chapter 5 Conclusions and future research ............................................................... 105 5.1 Conclusions ...................................................................................................... 105 5.2 Future research ................................................................................................. 107
References ....................................................................................................................... 112 Curriculum Vitae ............................................................................................................ 119
vi
List of Tables Table 3-1 LED structure of multiple DHs with sole SEI .................................................. 40 Table 3-2 LED structure with various active region designs and SEI thickness .............. 48 Table 3-3: PL decay times and amplitude ratios obtained from biexponential fits. ......... 80 Table 4-1: The ICP etching conditions ............................................................................. 93 Table 5-1 TLM pattern measurement results of delta doped p-GaN .............................. 109
vii
List of Figures Figure 1.1: LED lamps require less power to emit light than the older light sources. ....... 2 Figure 1.2: Globe LED market predicated by Yole Développement & EPIC .................... 3 Figure 1.3: The schematic diagram of GaN-based VCSEL structure with hybrid DBRs .. 6 Figure 1.4: The vertical profile of refractive index and standing optical wave in the ........ 7 Figure 2.1: Schematic of electron overflow caused by ballistic or quasi-ballistic electron transport across the InGaN active region. The electrons gain a kinetic energy after being injected into InGaN, which equals to E+ΔEc+qV(x). These hot electrons will either traverse the active region ballistically and quasi-ballistically, escape recombination inside InGaN, and contribute the electron overflow current, or be thermalized and captured inside the active region through interactions with LO-phonons. ...................................... 12 Figure 2.2: A schematic for the conduction band of a LED with a two-step layer SEI. (ΔEc 88meV) After being injected into the SEI from the n-GaN region, some electrons will have ballistic and quasi-ballistic, while the others (experiencing two or more scattering events) are considered to be thermalized in the SEI. ....................................... 14 Figure 2.3: Electron overflow percentile in (a) single DH LEDs and (b) multi-DH LEDs with different SEI thicknesses and active region designs as a function of injected current density. .............................................................................................................................. 16 Figure 2.4: Band structure of single 3 nm DH with 40 nm-thick GEI and SEI under current injection 100 A/cm2. GEI eliminates design requirements associated with the stepwise graded approach. Moreover, GEI provide more efficient electron cooling compared with two-step SEI with the same total electron injector thickness. .................. 18 Figure 2.5: Electron overflow percentiles calculated for (a) single and (b) quad 3 nm DH LED structures with either SEI or GEI of various thicknesses. ........................................ 19 Figure 2.6: Cross-sectional schematic of the InGaN-based DH samples investigated. The steps of different height are generated by ICP etching. .................................................... 21 Figure 2.7: Integrated PL intensities from underlying GaN, In0.01GaN layer, and the active region at 15 K and 295 K as a function of p-GaN thickness. The lines are exponential fits to the data for the active region. Data at 295 K and 15 K are shifted vertically for clarity. Error bars apply to all respective points for a given thickness. The inset shows a representative PL spectrum for the region with 100 nm thick p-GaN at 295 K. ....................................................................................................................................... 24 Figure 2.8: Electron flow percentile out of the active region into the adjacent n-type semiconductor of an InGaN LED used as a photodiode with photogenerated electron density to be the parameter. .............................................................................................. 25 Figure 2.9 Calculated IQE vs. current density for the case Auger term is figured in. The two sets of A, B and C values agree well with the LEDs from (a) Nichia and (b) Lumileds............................................................................................................................................ 29 Figure 2.10: With 425 nm excitation at 15 K, only the PL peak from low energy blue QW (460 nm) can be observed at an excitation density corresponding to 1019 cm-3 photo-generated carrier concentration. Note the lack of emission from the larger bandgap UV QW (400nm). The spectral range near the excitation laser line is blocked. ..................... 30 Figure 2.11: (a) Integrated PL intensities for the UV and blue QWs and the GaN layer in the two-color LED with 425 nm excitation at 15 K. (b) PL Intensity from the UV QWs versus the square root of the PL intensity from the blue QWs, which is proportional to the electron density in the blue QWs, nblue. ...................................................................... 31
viii
Figure 3.1: Device schematic of fabricated LED with contacts. ...................................... 35 Figure 3.2: (a) Schematic conduction band profile of coupled MQWs with HB with flat band for simplicity. (b) Relative EQE of MQWs LEDs as a function of pulsed injection current density (0.1 % duty cycle and 1 kHz frequency). ................................................. 36 Figure 3.3: Integrated PL intensity as a function of excitation power density at (a) 10 K and (b) 295 K; gray solid lines indicate slope of 1 and the inset of (b) displays the PL-IQE vs. the number of 3 nm DHs in the active region; (c) PL efficiencies of multi-3 nm DHs vs. excitation power density at room temperature. ................................................... 41 Figure 3.4: (a) The integrated EL intensity dependence on current density (the grey-sold line indicates slope of 1), (b) Relative EQE of multi-3 nm DHs vs. injected carrier density............................................................................................................................................ 43 Figure 3.5: Relative EQE of DH LEDs as a function of pulsed injection current density (0.1 % duty cycle and 1 kHz frequency)........................................................................... 45 Figure 3.6: The integrated PL intensity dependence on optical excitation density at (a) 15 K and (b) 295 K for single and quad 3 nm DH LEDs with varied SEI thickness. ........... 50 Figure 3.7: (a) The integrated EL intensity dependence on current density for DH LEDs with varied SEI thickness. (b) The relative EQE vs. injected current density. ................. 53 Figure 3.8: The schematics of hex 3 nm DH LED structures for the improvement of hole injections. Two 6 nm InGaN quantum barriers are located at the n-side for possible p-doping, while the other barriers are 3 nm. (a) In LED A, all the barriers are kept undoped. (b) In LED B, the first 6 nm barrier closest to n-side is p-doped. (c) In LED C, the two 6 nm barriers close to n-GaN were p-doped. ....................................................................... 57 Figure 3.9: (a) Electron concentration, (b) hole concentration, and (c) band structure) simulated for sample A (black), sample B (red) and sample C (blue) under current injection 100 A/cm2 with SILVACO ATLAS with parameters appropriate for nitride materials. ........................................................................................................................... 59 Figure 3.10: (a) The integrated EL intensities and (b) the relative EQE vs. injected current density for sample A, B, and C. ............................................................................ 62 Figure 3.11: (a) IQE of LEDs determined from the ratio of room temperature to low temperature PL intensities at the maximum excitation density employed, assuming unity internal quantum efficiency at 15 K. (b)The relative EQE vs. injected current density. .. 64 Figure 3.12: Energy band edge profiles simulated for (a) 3 nm DH and (b) 9 nm DH LEDs with 4+4 nm SEIs at different injection current densities. Peak emission energy shift as a function of injection current density from (c) EL measurements and (d) Silvaco simulations for DH LED structures with various active regions. ..................................... 67 Figure 3.13: Calculated coefficients Beff of (a) single DH, and (b) multi DH LEDs, calculated using squared overlap integrals of electron and hole wavefunctions (proportional to radiative recombination rate) within the active region as a function of current density using SILVACO ATLAS software package. The SEI layer thicknesses are provided in the legends in nm units. ................................................................................. 71
Figure 3.14: The PL decay time ( PL ), radiative decay ( R ) time and nonradiative
decay time ( NR ) for the hexa 1.5 nm (a), 2 nm (b) and 3 nm (c) LEDs as a function of
temperature ....................................................................................................................... 74
ix
Figure 3.15: (a) Cross-sectional SEM image of coalesced semi-polar GaN layers with 3 um x 3 um pattern; (b) inclined SEM image of non-coalesced semi-polar GaN layers with 3 um x 10 um pattern. ............................................................................................... 77 Figure 3.16: Steady-state room-temperature PL spectra of c-plane GaN films grown on sapphire and (1 100) m-plane and (1101)-oriented GaN layers grown on the Si patterned substrates. .......................................................................................................................... 78 Figure 3.17: Excitation density dependent room-temperature TRPL for (a) polar (0001) nano-ELO GaN film on sapphire and (b) semipolar non-coalesced semi-poar GaN layer on Si. ................................................................................................................................. 79 Figure 4.1: (a) Reflectivity of 29 pairs crack-free AlN/GaN DBRs on GaN template, (b) cross-sectional images of 29 pairs crack-free AlN/GaN DBRs from scanning electron microscope. ....................................................................................................................... 83 Figure 4.2: The structure schematic of the vertical cavity surface emitting laser: 2.5λ cavity with two hexa 3 nm In0.15Ga0.85N DHs separated by 132 nm (1λ) In0.01Ga0.99N underlying layer grown on bottom 29 pairs crack-free AlN/GaN DBRs on 4 µm GaN template and completed with 13 pair top SiO2/SiNx DBRs. ............................................. 84 Figure 4.3: The reflectivity spectrum (black) for the full vertical cavity, reflectivity spectrum for the bottom AlN/GaN DBR (red) and photoluminescence spectrum (blue). 85 Figure 4.4: Room temperature NSOM results for the microcavity structure with bottom semiconductor and top dielectric DBRs (a) height image (b) PL intensity mapping. ...... 86 Figure 4.5: The reflectivity spectrum of bottom 13 pair SiO2/SiNx bottom DBR. ........... 88 Figure 4.6: Cross-sectional SEM images of ELO-GaN grown (a) with constant NH3 flow for 3 hours and (b) with NH3 flow modulation for 2 hours. Wing tilt angle in case of NH3 flow modulation was measured to be lower than 0.1º whereas that in the case of constant NH3 flow was 5º. ............................................................................................................... 89 Figure 4.7: Cross section pictures of ELO growth for (a) ammonia on/off time 20 s/15 s with TMGa: 20 sccm and (b) ammonia on/off time 20 s/25 s with TMGa: 12 sccm ....... 91 Figure 4.8: The SEM cross-section image of ELO-GaN grown with pulse NH3 treatment............................................................................................................................................ 94 Figure 4.9: (a) cross-section SEM image and (b) top-view SEM image of ELO-GaN after etching with pyramids with etching condition I. .............................................................. 95 Figure 4.10: (a) cross-section SEM image and (b) top-view SEM image of ELO-GaN after etching with pyramids with etching condition II. ..................................................... 96 Figure 4.11: Cross-section SEM image I and (b) cross-section SEM image II of ELO-GaN after etching with pyramids with etching condition III. ........................................... 97 Figure 4.12: Cross-section SEM image and (b) top-view SEM image of ELO-GaN after etching with pyramids with etching condition IV. ............................................................ 97 Figure 4.13: (a) the VCSEL full vertical cavity structure on c-sapphire. (b) Electric field inside the cavity with respect to distance where an InGaN active region is placed at the antinode of the electric field inside the cavity. ................................................................. 99 Figure 4.14: Reflectivity spectrum of the full cavity structure. Cavity modes are observed around 400 and 412 nm can be clearly seen in inset. ...................................................... 101 Figure 4.15: PL spectra for the full cavity structure (red), half cavity structure (black) and reference sample (blue). .................................................................................................. 102 Figure 4.17: Cross-sectional schematics of fabricated VCSEL device with electrical contacts. .......................................................................................................................... 103
x
Figure 4.18: VCSEL devices on the mesa, (b) the final device shapes with contact layers.......................................................................................................................................... 104 Figure 5.1: delta- doping profile implementation: Stage I: undoped GaN growth (u-GaN); II: Nitridation; III: Mg incorporation (constant Mg molar flow for all the samples); .... 108 Figure 5.2: (a) Cross-sectional schematic of TLM pattern; (b) Measured hole concentration as a function of u-GaN spacer thickness .................................................. 110
Abstract
QUANTUM EFFICIENCY ENHANCEMENT FOR GAN
BASED LIGHT-EMITTING DIODES AND VERTICAL
CAVITY SURFACE-EMITTING LASERS
By
Fan Zhang
A research dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Electrical and Computer Engineering at Virginia Commonwealth University
Virginia Commonwealth University, 2014
Director: MORKOC, HADIS FOUNDER PROFESSOR OF ELECTRICAL AND COMPUTER ENGINEERING
This thesis explores the improvement of quantum efficiencies for InGaN/GaN
heterostructures and their applications in light-emitting diodes (LEDs) and vertical cavity
surface-emitting lasers (VCSELs). Different growth approaches and structural designs
were investigated to identify and address the major factors limiting the efficiency. (1) Hot
electron overflow and asymmetrical electron/hole injection were found to be the
dominant reasons for efficiency degradation in nitride LEDs at high injection; (2) delta p-
doped InGaN quantum barriers were employed to improve hole concentration inside the
active region and therefore improve hole injection without sacrificing the layer quality; (3)
InGaN active regions based on InGaN multiple double-heterostructures (DHs) were
developed to understand the electron and hole recombination mechanisms and achieve
high quantum efficiency and minimal efficiency droop at high injection; (4) the effect of
stair-case electron injectors (SEIs) has been investigated with different active region
designs and SEIs with optimized thickness greatly mitigated electron overflow without
sacrificing material quality of the active regions. The active regions showing promising
performance in LEDs were incorporated into VCSEL designs. Hybrid VCSEL structures
with bottom semiconductor AlN/GaN and a top dielectric SiO2/SiNx DBRs have been
investigated, and quality factors as high as 1300 have been demonstrated. Finally,
VCSEL structures with all dielectric DBRs have been realized by employing a novel
ELO-GaN growth method that allowed integration of a high quality InGaN cavity active
region with a dielectric bottom DBR without removal of the substrate while forming a
current aperture through the ideally dislocation-free region. The full-cavity structures
formed as such exhibited quality factors 500 across the wafer.
1
Chapter 1 Introduction
1.1 Motivation
In recent decade, III-nitride material has been extensively studied in application in
optoelectronic devices in the blue spectrum to the ultra-violet (UV), including light
emitting diodes (LEDs) and laser diodes (LDs)1, as well as applications in high frequency
and power electronics. GaN materials are very promising in power devices due to the
high breakdown field, high thermal conductivity, and high electron mobility (typically ~
1600 cm2/Vs at room temperature) in a typical AlGaN/GaN high electron mobility
transistors (HEMTs) grown by metal-organic chemical vapor deposition (MOCVD).
Moreover, tremendous improvements have been achieved in GaN-based LEDs.
InGaN/GaN blue LED, which boost the LED market, was the last -- and most difficult --
advance required to create white LED light. With white LED light, companies are able to
create smartphone and computer screens, as well as light bulbs that last longer and use
less electricity than any bulb invented before. The Nobel Prize in Physics 2014 was
awarded jointly to Dr. Isamu Akasaki, Dr. Hiroshi Amano and Dr. Shuji Nakamura for
the invention of efficient blue LEDs which has enabled bright and energy-saving white
light sources.
1.2 Development in GaN-based light-emitting diodes
GaN-based LEDs are used widely in various applications: liquid crystal display back
lighting source, general lighting source, sensing, and communications. Unlike
incandescent bulbs and fluorescent lamps, where most of the electricity is converted to
2
heat and only a small amount into light, GaN-based LEDs are capable of high efficiency
in conversion electricity to light. A typical solid state LED features a chip of
semiconductor heterostructures, i.e., InGaN/GaN, as a p-i-n junction. When electrons and
holes are injected into the junction under forward bias, they either recombine radiatively
to emit light or non-radiatively in the form of heating. GaN-based LEDs are more
efficient with higher luminous flux (measured in lumen) per unit consumed power
(measured in watt) compared to older light sources. The most recent record is just over
300 lumen/watt, which can be compared to 16 for regular light bulbs and close to 70 for
fluorescent lamps, 2 shown in Figure 1.1. As about one fourth of world electricity
consumption is used for lighting purposes, the highly energy-efficient LEDs contribute to
saving the resources consuming and significant reducing CO2 and SO2 gas pollution
generated from burning of fuel for electricity.
Figure 1.1: LED lamps require less power to emit light than the older light sources.
Although the cost this emerging technology is an issue to be addresses, one can virtually
certain that the cost of GaN related technologies will drop by orders of magnitude as this
3
technology matures. Yole Développement and EPIC announce their report dedicated to
the LED industry in Figure 1.2.
Figure 1.2: Globe LED market predicated by Yole Développement & EPIC As presented in Figure 1.2, the 2014 LED market size is estimated nearly $14.5 billion.
Among the applications, currently about 14% of total sale accounted for mobile
electronics, 21% for TV/monitors, and 48% for general lighting. It is expected that LEDs
for general lighting will grow further in the next few years. To 2018, the percentage will
be around 65% for the general light applications. Moreover, LEDs are constantly
improved. Recent results from Soraa Inc. indicated that the peak light output densities of
GaN-based LEDs exceeds 1000 W/cm2 and their external quantum efficiencies (EQE) are
maintained up to 56% at current density of 1000 A/cm2 without current crowding.3,4
However, challenges are also still prevailing in the process of LED market growth aside
of high cost issues including LED performance at high power levels.
4
One of the lingering problems for high power GaN-based LEDs is the efficiency
degradation, loss of efficiency at high electrical injection in LEDs beyond that which is
expected from thermally activated processes. The EQE reaches its peak value at current
densities as low as 50 A/cm2 (in some cases, around 5 A/cm2) and gradually decreases
with further increasing current injection5,6 Understanding the source of this degradation is
critical and minimization is desirable for the implementation of the GaN-based LEDs.
While the genesis of the efficiency degradation still remains a highly debatable topic, one
obvious mechanism is the asymmetry in doping in wide bandgap semiconductors, such as
GaN, wherein the hole injection falls behind that of electrons in the active region. The
radiative efficiency cannot keep up with increasing carrier injection due to progressively
lagging hole population in the active region. More importantly, Auger recombination
and/or carrier overflow are considered as the most probably major cause for the
efficiency degradation in InGaN LEDs. 7 , 8 , 9 , 10 Further discussion about asymmetry
electron/hole injection, Auger recombination and carrier overflow in InGaN/GaN active
region and how to improve the quantum efficiency of LEDs will be present in Chapter 2
and Chapter 3.
1.3 Development in GaN-based vertical cavity surface-emitting lasers
The vertical-cavity surface emitting laser (VCSEL) has several advantages over the
edge-emitting laser. Superiorities include the circular output beam, low beam divergence,
high modulation band-width, single longitudinal mode, and convenient wafer level
testing. 11 Owing to the geometry of this class of laser, VCSEL is promising for
optoelectronic devices for many applications, such as high-density optical storage system,
5
laser printing, and fiber-optic communications.12 Consequently, the VCSEL can combine
relatively low manufacturing costs with great performance.
However, in contrast to the impressive progress of GaN-based LEDs and edge-emitting
LDs, GaN-based VCSELs, which span 400 nm to 550 nm, still face significant
challenges.13 To understand challenges of GaN-based VCSELs, one must first understand
how this device operates. VCSEL is formed by sandwiched a thin active region between
two parallel mirrors, and the cavity length is of a few microns – hundreds of times shorter
than that of an edge emitting laser. The short cavity length enables high-speed direct
modulation. However, the gain per round-trip pass through the cavity is far less than that
for an edge-emitter. To compensate, cavity loss must be very small, and thus the
reflectivity of the mirror, which is a distributed Bragg reflector (DBR), must be very high
– it has to be nearly 99%. With a GaAs-based VCSEL, it is relatively easy to form such a
high reflectivity with a stack of lattice-matched, quarter-wavelength thick alternating
layers of GaAs/AlGaAs DBRs. Moreover, both GaAs and AlGaAs layers can be doped,
thereby enabling electrical injection into the cavity. Replicating this approach with the
III-Nitride growth has proved to be much hard. This had led several groups to introduce
new types of device structures, which either combine an epitaxial DBR with a dielectric
one, or employ dielectrics for both mirrors, which is shown in Figure 1.3.
6
Figure 1.3: The schematic diagram of GaN-based VCSEL structure with hybrid DBRs
Another consequence of the VCSEL’s short cavity length is the complex standing-wave
profile of the electric field intensity. To optimize laser performance, it is essential to
position the active region at the peak of this standing wave, while aligning lossy layers,
such as heavily doped layers, at a standing wave null. Very precise control of the cavity
length is needed to achieve this, shown in Figure 1.4. 14 This is readily achievable in all-
epitaxial structures, but more challenging in devices that feature double dielectric DBRs
and are formed with substrate lift-off or thinning.
7
Figure 1.4: The vertical profile of refractive index and standing optical wave in the center of the VCSELs
1.4 Scope of research
The research presented here concentrated primarily on GaN-based LED and VCSEL
structure designs to enhance the quantum efficiency.
The dissertation is organized as follows: Chapter 2 is a discussion of efficiency droop
issues persisted in the InGaN based LEDs. A review of the major causes will be included.
Numerical simulation with Silvaco Altas will be included for theoretical prediction of our
experiments. Our proposed mechanism will also be introduced and discussed.
Chapter 3 focuses on how to mitigate efficiency droop and improve quantum efficiency
for GaN-based LEDs. Various structure designs will be presented and their effects on
quantum efficiency will be discussed. Under this realm, the comparison of LED
performance with/without delta p-doped barrier and with high or low height barrier to
provide a understanding on efficiency limitation caused by poor hole transport. We will
also present with multiple DH active region design and how the EQE was improved by
8
incorporating more thin DH in the active region. Finally we presented that with optimized
SEI for a given active region, the EQE can be further improved.
Chapter 4 focuses on quantum efficiency of GaN-based VCSELs. Various structure
designs will be presented and their effects on quantum efficiency will be discussed.
Under this realm, quality limitation of hybrid VCSEL will be discussed. We will also
present the novel growth technique, fabrication process and measurement of VCSEL with
both dielectric DBRs.
Finally conclusions and suggestions for future work will be presented in Chapter 5.
9
Chapter 2 Efficiency droop investigations
2.1 Introduction to the efficiency droop
InGaN based light emitting diodes (LEDs) are becoming widely used for general
lighting and displays with internal quantum efficiency (IQE) and optical extraction
efficiency in high performance devices being in the range of 80%.15 High power blue
InGaN based LEDs are now able to produce over 250 lm/W efficiency with wall plug
efficiencies, optical power divided by the electrical power, in excess of 60% through
improved quality and layer structural design, which is simply remarkable. However,
LEDs suffer from the reduction of efficiency at high injection current levels. As shown in
Figure 2.1 , efficiency drops by about 50% as current density increases from a few
amperes per square centimeter to a few hundred amperes per square centimeter, which
greatly affects the further application of LEDs. The DoE's office of Energy Efficiency
and Renewable Energy lists it as a top technical issue to be resolved in solid-state
lighting.16
10
Figure 2.1 Efficiency droop in a typical blue LED, shown in relative units17
To account for the efficiency droop, various models have been proposed, including
current roll-over 18 , poor hole injection 19 , 20 , 21 , polarization field 22 , 23 , Auger
recombination24, and junction heating25. Weisbuch et al. reported detection of Auger
recombination in LEDs from the existence of Auger electrons by electron emission
spectroscopy,26 but it is unclear whether Auger recombination is the dominate cause of
efficiency degradation or not. Moreover, the Auger loss in wide band-gap semiconductor
as GaN is expected to be small, as verified by fully microscopic many body models.27,28
In addition, if Auger recombination was solely responsible for the efficiency degradation,
GaN-based lasers, requiring high injection levels, would be undoubtedly prevented,
which is not the case.29 So far, the extensive experimental results suggest that the
efficiency droop is related to the skewed carrier injection due to the disparity of hole and
electron concentrations, large hole effective mass, and carrier overflow instead of the
Auger recombination. As reported, Xie et al. employed either p-type doped InGaN
11
barriers or lightly n-type doped GaN electron injection layer just below the MQWs to
achieve comparable levels of electron and hole injection and observed mitigation of the
droop, which suggest poor hole transport and injection through the barrier being the
responsible mechanism.30 Moreover, Ni et al. offered further supporting data for the
electron overflow by investigating the efficiency droop in double heterostructure (DH)
LEDs with different hot electron stopper or cooler designs both theoretically and
experimentally.31
Here we propose the electron overflow and lagged hole injection to be the dominant
mechanism responsible for the efficiency degradation. The term “overflow electrons”
refers to the electrons which escape the active region without participating in any
recombination process, and end up recombining in the p-GaN region or make it to the p-
contact if the minority carrier lifetime in that region permits it. We will investigate
possible solutions to eliminate the electron overflow via numerical simulation in this
chapter. The discussion of improve hole injection of LEDs will be discussed in Chapter 3.
2.2 Hot electron model
In previous work, Ni et al. evaluated non-equilibrium electrons, i.e. hot electrons, inside
in the InGaN/GaN LEDs. 32 The injected electrons, which acquire additional kinetic
energy equal to conduction band offset when entering active region, can traverse the
active layer by ballistic or quasi-ballistic transport without recombination and recombine
in the p-GaN region. We should note that the nonradiative recombination is prevalent in
p-type GaN due to extremely long lifetime. Dr. Ni used first-order estimation of the hot
electron effect and explained our experimental data with varying barrier height of the
12
EBL. The calculation assumes that the electrons obey the Fermi-Dirac distribution in the
n-GaN layer before they are injected into the active region. The electrons acquire the
additional kinetic energy equal to the conduction band offset between n-GaN and InGaN
active region (In0.15Ga0.85N active region is typical used for LEDs in our experiments)
upon injection. These hot electrons would either undergo thermalization and lose their
excess energy mainly through interaction with LO-phonons33 or avoid thermalization and
escape the InGaN region as depicted in Figure 2.1.
Figure 2.1: Schematic of electron overflow caused by ballistic or quasi-ballistic electron transport across the InGaN active region. The electrons gain a kinetic energy after being injected into InGaN, which equals to E+ΔEc+qV(x). These hot electrons will either traverse the active region ballistically and quasi-ballistically, escape recombination inside InGaN, and contribute the electron overflow current, or be thermalized and captured inside the active region through interactions with LO-phonons. The calculations on the hot electron overflow took into account the ballistic electrons,
representing those that experience no scattering in the active region, and the quasi-
ballistic electrons that experience one scattering event (i.e. quasi-ballistic motion
involving either one LO-phonon emission or absorption), and two scattering events (4
13
combinations of two scattering events involving LO-phonon emission and absorption).
Those experiencing multiple energy loosing scattering events are eventually thermalized.
In the calculations the electrons are categorized according to their scattering events that
they experience: no scattering, one scattering event (one LO-phonon emission or one LO-
Phonon absorption), etc. the calculated contribution of electrons undergoing two
scattering events to the overflow is less than 1% of the total injected electrons and can be
neglected in our experimental device structures. Thus, based on the following three
scattering events: (1) – no scattering, (2) – one phonon emission, (3) – one phonon
absorption, the hot electron overflow can be estimated.
2.3 Stair-case injector (SEI) designs
In this section, we will concentrate on the designs of the SEI structures for optimum
impact. Theoretically, a sufficiently large step height and a larger SEI thickness will
reduce the electron overflow (an optimum step height for the SEI would provide a
balance between the gained electron kinetic energy in the staircase region and the
overflow contribution from the electrons thermalized within the SEI). However, due to
the growth technical limitations, thicker InGaN SEI layers have better chance to be
relaxed and thus more threading dislocations could be generated and penetrated into the
active region leading to the much increased non-radiative recombination. Therefore,
growth related issues should also be taken into consideration when optimizing the SEI
layer stack, which will be discussed in details later. The schematic structure is shown in
Figure 2.2.
14
Figure 2.2: A schematic for the conduction band of a LED with a two-step layer SEI. (ΔEc 88meV) After being injected into the SEI from the n-GaN region, some electrons will have ballistic and quasi-ballistic, while the others (experiencing two or more scattering events) are considered to be thermalized in the SEI. A universal solution in the InGaN based LED structures in commercial companies to
prevent electron overflow is to employ p-AlGaN electron blocking layer (EBL). However,
it should be noted that the EBL impedes hole injection due to the generated valence band
potential barrier between active region and p-GaN. Moreover, in the commercial c-plane
GaN based LEDs, the AlGaN EBL is located on top of the InGaN barrier of the active
region. The lattice mismatch between AlGaN and InGaN layers generates piezoelectric
polarization field in addition to differential spontaneous polarization fields, which pull
down the conduction band at the AlGaN/InGaN interface. As a result, the effective
barrier height of the AlGaN EBL is a compromise, and the electron overflow is not
effectively suppressed.34 InGaN SEI, inserted between n-GaN and InGaN active region,
can reduce the electron overflow by gradually thermalizing down the hot electrons
without blocking hole injection. It was demonstrated experimentally that with the
insertion of SEI, the efficiency degradation of LEDs could be substrantially reduced and
15
the AlGaN EBL, employed to reduce electrons overflow but hampers hole transport,
could be safely replaced with the SEI.35 LED structures grown on non-polar m-plane bulk
GaN with 6-nm DH active region show almost identical dependence of EQE on current
density when they contain either a three-layer SEI only or both the SEI and an EBL,
while the first-order calculations of the electron overflow at different applied forward
voltage lead to a high overflow percentage from 60% to 90% for the LEDs without any
SEI and EBL and this percentage can be significantly reduced down to 10-20% by
inserting the SEI only. Furthermore, the first order calculation for the c-plane 6-nm DH
LEDs reveals that the electron overflow for the LED with one-layer SEI (no EBL)
saturates at ~18% once the applied voltage exceeds 6 V, while the overflow increases
with the applied voltage from 0% to 50% when the voltage increases from 3 to 14 V for
the LED with EBL if no SEI is incorporated. Therefore, most of the investigated LED
structures in this thesis have incorporated SEI structures without employment of EBL
unless indicated specifically.
It should be noted that the specifics of the SEI region would depend on the specifics of
the active region, thicker active regions relaxing the design parameters. In this context,
the electron overflow vs. injection current for a variety of single and multi In0.15Ga0.85N
DH LEDs with two-layer SEIs of different thicknesses (with no electron blocking layer)
has been calculated. For simplicity, the electrons are assumed to move in the direction
normal to the hetero-interfaces only and the total electron overflow is obtained by
summing the contributions from both ballistic and quasi-ballistic electrons for a given
bias voltage. The injected current density vs. bias was obtained using a commercial
16
simulator, SILVACO ATLAS, with parameters appropriate for nitride materials as
reported in Ref.35.
0 200 400
0
20
40
60
80
100
0 200 400
Ele
ctro
n O
ver
flow
Per
centi
le (
%)
(b) (a)
Ele
ctro
n O
ver
flow
Per
centi
le (
%)
Current density (A/cm2)
3nm SEI 20+20nm 6nm SEI 20+20nm 9nm SEI 20+20nm
3nm SEI 4+4nm 6nm SEI 4+4nm 9nm SEI 4+4nm
3nm No SEI 6nm No SEI 9nm No SEI
Dual 6nm no SEI Quad 3nm no SEI Hexa 3nm no SEI Quad 3nm GaN barrier no SEI
Dual 6nm SEI 4+4nm Quad 3nm SEI 4+4nm Hexa 3nm SEI 4+4nm Quad 3nm GaN barrier SEI 4+4nm Quad 3nm SEI 20+20nm
Figure 2.3: Electron overflow percentile in (a) single DH LEDs and (b) multi-DH LEDs with different SEI thicknesses and active region designs as a function of injected current density.
Figure 2.3 shows the schematic and the computed electron overflow percentiles for two
types of LED structures (no EBL involved): (1) single DH LEDs with In0.15Ga0.85N
active-region width varying from 3 to 9 nm and (2) multi-DH LEDs with different
number of In0.15Ga0.85N active regions separated by 3 nm In0.06Ga0.94N barriers or GaN
barriers. As evident from Figure 2.3 (a), the rate of the electron overflow reduces with
increasing active region width. In the absence of an SEI, the overflow percentile for the
single 3 nm DH LEDs reaches 68% at ~75 A/cm2, while only 28% of electrons escape to
the p-GaN in the 9 nm DH structure. At 150 A/cm2 (injection current density beyond that
used in general-lighting LEDs), the electron overflow reduces from 75% in 3 nm to 45%
and 32% for the 6 nm and the 9 nm DH structures, respectively. These values are reduced
17
to 31%, 16%, and 10% for the 3 nm, the 6 nm, and the 9 nm DHs, respectively, when a
4+4 nm-thick two-layer SEI is employed and further to ~5% with a 20+20 nm thick SEI.
It should be noted, however, that although the carrier overflow reduces with increasing
DH thickness, the use of DH active regions with thickness beyond 6 nm may not be
beneficial, partly because nonradiative recombination rate increases with increasing
active region thickness beyond certain thickness due to material quality degradation.
Furthermore, too wide active region would substantially reduce radiative recombination
efficiency because of poor overlap of electron and hole wavefunctions in wide active
regions. Actually it is promising to feature structures like multiple quantum wells
(MQWs) or DHs with thin active layers but cumulatively sufficiently thick active regions.
Figure 2.3 (b) illustrates that the thick cumulative active region naturally provides
efficient electron cooling. Moreover, when the band offset between n-In0.01Ga0.99N
underlying layer and barriers in the active region becomes larger than the LO phonon
energy of 88 meV, the electrons are thermalized efficiently because the whole active
region inclusive of the In0.06Ga0.94N barriers works naturally as an electron cooler. Owing
to the relatively small conduction band discontinuity between the In0.06Ga0.94N barriers
and the In0.15Ga0.85N active layers, the reduced barrier heights not simply contribute to the
electron thermalization and lower the velocity of electrons entering the active regions, but
also benefit hole transport in the active region. As evident from Figure 2.4 (b), the
electron overflow percentile in the quad 3 nm DH without SEI or EBL reduces from 28%
to 16% at 75 A/cm2 when In0.06Ga0.94N is used as the barrier material instead of GaN and
from 56% to 31% at 500 A/cm2.
18
2.4 Graded Electron Injector (GEI)
Graded electron injector (GEI) is using a stepwise InGaN electron cooling layer with
increasing In composition gradually, compared with two-step staircase electron injector.
The conditions imposed on heterojunction discontinuities in the case of SEI are
eliminated entirely as the structure automatically cools electrons when they have
sufficient potential energy to emit LO phonons. Moreover, electrons in the graded
injector are continuously accelerated in the direction normal to the heterointerface due to
the electric field and more efficient electron cooling is provided compared with SEI with
the same total electron injector thickness, shown in Figure 2.4. Furthermore, as in the
case of the SEI design, the total thickness along with the active layer would have to be
considered in the realm of strain and strain relaxation with its ensuing potential defect
generation.
-0.2
0.0
0.2
0.4
0.00 0.02 0.04 0.06
-0.2
0.0
0.2
0.4
(a)
(b)
Growth direction
single 3nm DH with GEI 40nm
single 3nm DH with SEI 20+20nm
Conduct
ion B
and (
eV
)
Distance (m)
Figure 2.4: Band structure of single 3 nm DH with 40 nm-thick GEI and SEI under current injection 100 A/cm2. GEI eliminates design requirements associated with the stepwise graded approach. Moreover, GEI provide more efficient electron
19
cooling compared with two-step SEI with the same total electron injector thickness.
0 100 200 300 400 500
10
20
30
40
0 100 200 300 400 500
10
20
30
40
Ele
ctro
n o
ve
rflo
w p
en
ce
nta
ge
(%
)
Current density (A/cm2)
Ele
ctr
on
ove
rflo
w p
en
ce
nta
ge
(%
)
3nm DH SEI 4+4nm 3 nm DH GEI 8nm 3nm DH SEI 20+20nm 3 nm DH GEI 40nm
4x3 nm DH SEI 4+4nm 4x3 nm DH GEI 8nm 4x3 nm DH SEI 20+20nm 4x3 nm DH GEI 40nm
Figure 2.5: Electron overflow percentiles calculated for (a) single and (b) quad 3 nm DH LED structures with either SEI or GEI of various thicknesses.
Figure 2.5 compares the electron overflow percentiles calculated for single and quad 3
nm DH LED structures with either SEI or GEI. It is clear that the elimination of band
discontinuities in the GEI reduces the electron overflow considerably. For single 3 nm
DH LEDs, the overflow at 100 A/cm2 is reduced from 26 % with 4+4 nm thick SEI to 12
% with an 8 nm thick GEI. These values are reduced to 12 % and 10 % for 20+20 nm SEI
and 40 nm GEI, respectively. For quad 3 nm LEDs, where the thicker active region
naturally contributes to electron cooling, the overflow at 100 A/cm2 is reduced from 10 %
with 4+4 nm SEI to 7 % with an 8 nm thick GEI. The corresponding values are 5.5 % and
3.5 %, respectively, for quad 3 nm LEDs with 20 + 20 nm SEI and 40 nm GEI.
20
2.5 Electron leakage from active region
The electron overflow in forward biased LEDs must be delineated from the carriers’
recombination, including radiative and nonradiative recombination, by hot electron
models and thus somewhat indirect. In order to have a clear picture of electron overflow,
we measured electron leakage from the active region in reversed biased LEDs using the
photodiode configuration with optical excitation, where electrons were traversing to the
n-GaN of LEDs. The portion of the generated electron-hole pairs that participates in
radiative recombination can be estimated by luminescence emanating from the junction.
The contribution by the nonradiative component can be neglected as the layer quality
nowadays is sufficiently good to justify this assumption. In the experiments we
undertook, CW 325 nm HeCd excitation had to be used which is absorbed by the p-layer
and the fraction of photons absorbed in the active region can only be determined if the
minority carrier diffusion length is known.
2.5.1 Diffusion length measurement in p-GaN
First, we must determine the diffusion length by progressively thinning the p-layer,
which are then used in the estimation of the photons that are absorbed in the active region
in conjunction with the photodiode experiments. Here, we investigated diffusion lengths
of the photo-generated carriers in p-type GaN along the c-direction, using
photoluminescence (PL) spectroscopy, which do not require advanced nonlinear optical
techniques or electrical contacts as for most of the abovementioned methods.
21
The samples used were c-plane 6 nm thick In0.15Ga0.85N double heterostructure (DH)
active regions grown on a ~3.7 µm-thick n-type GaN template on sapphire in a vertical
low-pressure metalorganic chemical vapor deposition (MOCVD) system. A 60 nm Si-
doped (2×1018 cm-3) In0.01Ga0.99N underlying layer was inserted just beneath the active
region for improving the quality of the overgrown layers. The structures were completed
with either Mg-doped (chemical doping concentration ~ 1019 cm-3) p-GaN layer of 500-
nm thickness, shown in Figure 2.6. Hole concentration in p-GaN was determined to be
4×1017 cm-3 from Hall measurements on a separate calibration sample. To mitigate the
effect of any Mg out-diffusion from p-GaN on the optical quality of the active region, a
20 nm thick In0.01Ga0.99N spacer layer was grown in between.
Figure 2.6: Cross-sectional schematic of the InGaN-based DH samples investigated. The steps of different height are generated by ICP etching.
To achieve p-type layers with different thickness, selective area inductively coupled
plasma (ICP) etching was used in multiple steps. ICP etching introduces surface damage
22
which modifies the PL intensity. To minimize this effect, a two-step etching procedure, a
high power, i.e. physical etch step followed by a low power, i.e. chemical etch, was
employed. This two-step etching process was found to help with the surface recovery so
that the effect of etching on PL intensity could be neglected. Etched thickness of the top
layer was determined after each etching step using a surface profiler. PL measurements
were carried out on regions with different p-GaN thicknesses using He-Cd laser (325 nm
wavelength) excitation with resulting photogenerated carrier concentrations (n) from
1.5×1018 cm-3 to 4.2×1018 cm-3. At a given excitation power, the carrier density was
determined from the generation rate at steady state, 2 3G An Bn Cn , assuming that
electron and hole capture cross-sections are equal, and n is much higher than the
background carrier concentration, which is justified (background concentration ~1017 cm-
3). The Shockley-Read-Hall, bimolecular, and Auger recombination coefficients of A =
107 s-1, B = 5×10-11 cm3s-1, and C = 10-31 cm6s-1, respectively, were used.36 The carrier
generation rate, G, was calculated using (1 )G P R Sh , where P is the excitation
laser power incident on the sample, α is the absorption coefficient at the excitation photon
energy hν, R is the Fresnel reflection coefficient for the sample/air surface, S is the laser
spot size on the sample surface. Due to the large absorption coefficient (1.1×105 cm-1)37
at the excitation wavelength, carrier photogeneration takes place near the surface region
of the sample, and upon diffusion of carriers away from the surface the PL emanating
from the active region, spacer layer, underlying InGaN and underlying n-GaN is
measured. The thicker the top layer, the less intense the PL is from the underlying layers.
The PL intensity of the etched region was compared with that from the corresponding un-
etched reference region to minimize the effect of any variation across the sample. It
23
should be noted that as the top GaN layer is made thinner, more of the excitation light
penetrates into the active region and the underlying layers and gets absorbed there. The
resulting PL is then due to recombination of both the directly photogenerated carriers and
those diffusing from the top GaN.
Figure 2.7 shows the integrated PL intensity of the active region plotted as a function of
the p-GaN layer thickness at 295 K and at 15 K, taking into account the absorption of
incident laser power in the top p-GaN and In0.01GaN spacer layer. With increasing p-GaN
layer thickness, the normalized PL intensity exhibits an exponential decay of the
form , where x is the p-GaN thickness and Ldiff is the diffusion length. The
diffusion lengths in p-GaN extracted from the fits are 93±7 nm and 70±7 nm at 295 K
and 15 K, respectively. The error in the diffusion length measurements by PL originates
mainly from the thickness nonuniformity across the sample introduced by multiple ICP
etching steps. A typical PL spectrum (region with 100 nm thick p-GaN) is shown in the
inset of Fig. 2. As is often the case, it should be noted that due to high Mg doping, there
is nearly no near band edge emission from the p-GaN layer.
24
0 100 200 300 400 500
360 390 420 450 480
p-GaN thickness (nm)
Norm
aliz
ed
in
tegra
ted
PL
295 K Underlying n-GaN In
0.01GaN layer
Active region
15 K Underlying n-GaN In
0.01GaN layer
Active region
active region
n-GaN
In0.01GaN
PL
in
ten
sity
(a
.u.)
Wavelength (nm)
Figure 2.7: Integrated PL intensities from underlying GaN, In0.01GaN layer, and the active region at 15 K and 295 K as a function of p-GaN thickness. The lines are exponential fits to the data for the active region. Data at 295 K and 15 K are shifted vertically for clarity. Error bars apply to all respective points for a given thickness. The inset shows a representative PL spectrum for the region with 100 nm thick p-GaN at 295 K.
The decrease in diffusion length at low temperature in p-GaN is mainly due to the
increased ionized impurity scattering which is dominant at low temperatures. As the
thermal velocity of the carriers reduce, the effect of long-range Coulomb interactions on
their motion increases. Moreover, radiative recombination rate increases with decreasing
temperature, and consequently, shorter recombination lifetime reduces the diffusion
length. This temperature dependence of diffusion length is also consistent with data
available in literature.38
25
2.5.2 Electron escape from active region
-3 -2 -1 0
0.12
0.14
0.16
0.18
0.20
n (cm-3
) 9.3x1017
7x1017
3.7x1017
2.3x1017
9x1016
Esc
aped
to
gene
rate
d ca
rrie
r ra
tio
Voltage (V)
Figure 2.8: Electron flow percentile out of the active region into the adjacent n-type semiconductor of an InGaN LED used as a photodiode with photogenerated electron density to be the parameter.
Figure 2.8Error! Reference source not found. shows the percentile of electron
leakage out of the active region vs. the bias for an LED containing a 6 nm-thick
In0.15Ga0.85N active layer and a 10 nm-thick Al0.20Ga0.80N EBL. Naturally an increase
with bias ensues because of the band bending due to the bias favoring electron escape
from the active region. LEDs and photodiodes are dissimilar in that in the former one
desire to have all the carriers to recombine in the active region and the recombination
current to the only current component. In the latter one desires to extract all the
photogenerated carriers out of the absorption region which means that a well-designed
LED would not make a very good photodetector but for the sake of the experiments
discussed here, LEDs can be used in the photodiode realm. The photocurrent method can
26
be used to estimate the carrier overflow in LEDs by investigating structures with barriers
of different height inserted on both sides of the active region.
2.6 Carrier overflow versus Auger recombination
Using the assumption that the recombination region is much thinner than the natural
diffusion length of minority carriers for DH or MQW LEDs, the recombination rate
equation under steady state is typically described as 2 3 /An Bn Cn J qd . The term C
represents Auger nonradiative coefficients. Therefore, in the absence of carrier spillover
the IQE can be written as2 2 3
int / / ( )eff r Bn An Bn Cn . If it were possible to
measure IQE vs. the injected current density J, one could obtain A, B, and C coefficients
through a third order polynomial fitting, albeit without a unique solution. The collected
radiative power can also be measured vs. the optical excitation power from which A and
C coefficients can be deduced for a given B coefficient assuming that the extraction and
collection efficiency remains the same for all excitation levels39. Let us now turn our
attention to calculating A and B coefficients first. Assuming a trap density of Nt ≈ 1016
cm-3, a capture cross section of σ = 10-15 cm2 for this particular trap, and a thermal
velocity of vth = 5×106 cm/s, which are reasonable, A coefficient is found to be
1/ nr th tA N = 5×107s-1, which is consistent with other report40 where this coefficient
is extracted from a fitting of the light output dependence on the injection current density.
For a detailed calculation of the B coefficient, one can refer to Ref. 30, but in a simple
sense, B = G/ni2 ≈ 4.1×10-9 cm3s-1 for GaN, where G is the generation rate per unit
volume and ni is the intrinsic carrier concentration. Doing so leads to a B coefficient
value for In0.2Ga0.8N of approximately 1.2×10-10 cm3s-1. These calculated A and B values
27
will serve as reference ranges for those extracted from fits to the curves of the light
output vs. injection current density in order to reduce the error in values obtained from
the polynomial fitting.
According to Ref. 41 and as shown in Figure 2.9, the calculated C values for both
LEDs are already out of the expected range of 1.4×10-30 cm6s-1 to 2×10-30 cm6s-1
predicted in Ref. 24.
28
0 20 40 60 80 1000.0
0.2
0.4
0.6
0.8
1.0
Exp. data - Nichia LED Fit
A = 1x107s
-1
B = 1.9x10-10
cm3s
-1
C = 3.2x10-29
cm6s
-1
IQE
Current density (A/cm2)
(a)
0 100 200 300 400 5000.0
0.2
0.4
0.6
0.8
1.0
Lumileds
A = 5.4x107s
-1 , B = 2x10
-11cm
3s
-1, C = 2x10
-30cm
6s
-1
Fit (fixed B = 1.9x10-10
cm3s
-1)
A = 1.75x108s
-1
C = 5.5x10-29
cm6s
-1
IQE
Current density (A/cm2)
(b)
29
Figure 2.9 Calculated IQE vs. current density for the case Auger term is figured in. The two sets of A, B and C values agree well with the LEDs from (a) Nichia and (b) Lumileds.
Returning to more conventional experiments, there are also other observations to
indentify effect of Auger recombination in carrier recombination. Binder et al. 42
investigated two color structures wherein when the longer wavelength layer (green) is
resonantly excited, the shorter wavelength (UV) region emits light, which was attributed
to hot electrons created by the Auger process in the longer wavelength layer (green)
traversing to the shorter wavelength one (UV) followed by radiative recombination. This
report, however, is void of quantum well (QW) thicknesses and excitation power density
levels used. Moreover, Hader et al.,43 using a fully microscopic model to emulate this
particular experiment, argue that resonant laser tuned to create carriers only in the green
region does also create carriers in the UV region even in the absence of any Auger
process. The reason provided is that the polarization excited by the spectrally narrow
optical pulse dephases in few tens of femtoseconds due to electron-electron and electron-
phonon scattering. In this scenario, the free carriers are generated through a coupling
between the optical pulse and the associated material polarization causing the total
spectral width of the excitation to be dominated by dephasing time of the polarization,
not by that of the optical pulse. The resulting spectral width is argued to be sufficiently
wide to excite the UV QW as well. Due to the very small coupling between the states in
the green region and those of the UV region, the carriers generated in the UV region do
not relax to states in the green region, which paves the way for UV emission. In our own
experiments dealing with two color systems, the resonant excitation of the blue QW
region using a frequency doubled Ti:Sapphire laser of ~ 100 fs pulsewidth and 80 MHz
30
repetition rate did not result in any detectable emission from the UV QWs (400 nm) up
the maximum laser excitation density that can be applied in our lab which corresponds to
a carrier density of 1019 cm-3, much larger than the injection densities employed in LEDs,
as can been seen in Figure 2.10. It should also be noted that this carrier density is
estimated using the steady-state approximation, which certainly is not fully valid for the
pulsed excitation conditions used, and therefore, represents an average value and only a
lower limit of the instant carrier density within the active region.
380 400 420 440 460 480 500 520
No PL is observed from
UV QW
Wavelength (nm)
PL
Int
ensi
ty (
a.u.
)
Laserexcitation(425 nm)
PL from blue QW
Figure 2.10: With 425 nm excitation at 15 K, only the PL peak from low energy blue QW (460 nm) can be observed at an excitation density corresponding to 1019 cm-3 photo-generated carrier concentration. Note the lack of emission from the larger bandgap UV QW (400nm). The spectral range near the excitation laser line is blocked.
When the two-color LED mentioned above was excited using a 1 kHz repetition rate 130
fs pulsewidth optical parametric amplifier (425 nm wavelength) providing much higher
photon fluences allowing clear observation of two-photon absorption related emission
from all layers, still no evidence of Auger recombination related transfer of hot electrons
from the blue QW to the UV QW was observed. Error! Reference source not found.
31
2.11 (a) shows the PL intensities for the UV and blue QWs as well as the underlying GaN
layers as a function of the excitation density. Under the excitation conditions used the PL
intensities from the UV and blue QWs are proportional to 2UVBn and 2
blueBn , where nUV
and nblue are the electron densities in the UV and blue QWs, respectively. If Auger
recombination were effective in the blue QWs, their generation rate ( 3blueCn ) would be
equal to the generation rate in the UV QWs ( 2UVBn under steady-state) following hot
electron transport from the blue QWs under the assumption that Auger recombination in
the UV QWs and the SRH recombination overall at 15 K are negligible. This would then
imply that the intensity from the UV QWs should exhibit cubic dependence on the square
root of the intensity from the blue QWs. However, the observed dependence above the
threshold for the UV QW emission is to the 6th power as shown in Figure 2.11 (b).
101
102
103
Inte
gra
ted P
L Inte
nsi
ty (
arb
. un
its)
Excitation energy density (J/cm2)
blue QW UV QW GaN
15 K
(a)
slope 2
UV
QW
PL Inte
nsi
ty (
arb
. units
)
(blue QW PL Intensity)1/2
(arb. units)
(b)
slope 6
Figure 2.11: (a) Integrated PL intensities for the UV and blue QWs and the GaN layer in the two-color LED with 425 nm excitation at 15 K. (b) PL Intensity from the UV QWs versus the square root of the PL intensity from the blue QWs, which is proportional to the electron density in the blue QWs, nblue.
32
The observed quadratic dependence of the UV QW PL on excitation density in Figure
2.11 (a) suggests that the UV emission is due to generation of carriers in the UV QWs by
two-photon absorption. The same quadratic dependence is observed for the PL from the
underlying GaN layer. It can, therefore, be concluded that the Auger recombination is not
effective at carrier densities that correspond to the threshold for UV emission, beyond
which significant two-photon absorption occurs. The average carrier density estimated
for the threshold excitation density value of 100 J/cm2 is 2 x 1019 cm-3, which, as
indicated above, represents only a lower limit due to the femtosecond pulsed excitation
used.
33
Chapter 3 Quantum efficiency enhancement in GaN-based LEDs
3.1 Effects of quantum barrier: height and thickness
3.1.1 Motivation
We have undertaken a series of theoretical investigations to unveil the root cause of the
dominant efficiency degradation mechanism: hot electron overflow and lagged hole
injection compared with electron injection. In order to retain the quantum efficiency at
high injection, one approach would increase of the number of quantum wells (QWs) in
the active region. However, due to the poor hole transport in InGaN LEDs, light emission
is attributed mainly to the QWs closest to the p-GaN layer in typical c-plane GaN-based
LEDs. Another approach is to utilize double heterostructure (DH) active regions for
uniform carrier spreading across the active region. In this realm, different barrier height
(either In0.01GaN or In0.06GaN barriers) and thickness (3 nm and 12 nm) applied on
MQWs were investigated.
3.1.2 Experimental procedures
The c-plane InGaN LED structures, emitting at ~420 nm, were grown on ~3.7 µm-thick
n-type GaN templates on sapphire in a vertical low-pressure metalorganic chemical vapor
deposition (MOCVD) system. The structures feature various active regions such as
variants of MQWs and DHs, with two-step SEI (two 4-nm InGaN layers with step-
increased Indium composition of 4% and 8%) for optimization of internal and external
quantum efficiencies. The SEI steps have conduction band potential energy drop equal to
or more than one LO-phonon energy (92 meV for GaN) and enhance electron
thermalization through LO-phonon emission. The MQWs active regions are with
34
different barrier height (either In0.01Ga0.99N or In0.06Ga0.94N barriers) and thickness (3 nm
and 12 nm). All the structures contained a 60-nm Si-doped (2×1018 cm-3) In0.01Ga0.99N
underlying layer below the active region and SEI for quality improvement. The LED
structures were completed with 100 nm-thick Mg-doped p-GaN layers having 4×1017 cm-
3 hole density, determined by Hall measurements on a separate calibration sample. For
devices, square mesa patterns (400x400 μm2) were formed by conventional lithography
and chlorine based Inductively Coupled Plasma (ICP) etching. Ti/Al/Ni/Au
(30/100/40/50 nm) metallization annealed at 800 ºC for 60 seconds was used for n-type
ohmic contacts, and 5 nm/5nm Ni/Au electrodes served as the semi-transparent p-contact,
with 40/50-nm Ni/Au electrodes deposited for p-contact pads. The fabricated device is
shown in Figure 3.1.
35
Figure 3.1: Device schematic of fabricated LED with contacts.
3.1.3 Results and discussion
The relative EQE is measured on-wafer pulsed EL measurements (0.1 % duty
cycle, 1 kHz) carried out for all the investigated LEDs with no effort having been made
to enhance light extraction. Figure 1 (a) shows the schematic for the LED structures with
coupled-MQW In0.01Ga0.99N high barrier (HB). Figure 1(b) shows the relative EQE
values for the MQW-LED structures. Among them, the relative EQE for the low barrier
(LB) MQWs is always higher than MQWs with high barriers (30% in coupled-MQWs
series and 15% in uncoupled-MQWs series). It is well known that the hole transport in
GaN is compromised due to the large hole effective mass and ensuing low hole mobility.
The lower In0.06GaN barrier height in the MQW active regions would help the hole
transport and quantum tunneling ascribed to the lowered barrier and thus reduce the
electron overflow induced efficiency degradation at high injection levels as the
probability of recombination in the active region would increase with more holes present.
36
0 100 200 300 400 500 600
Current density (A/cm2)
Re
lativ
e E
QE
(b)
Uncoupled_MQW_LB Uncouple_MQW_HB Coupled_MQW_LB Couple_MQW_HB
Figure 3.2: (a) Schematic conduction band profile of coupled MQWs with HB with flat band for simplicity. (b) Relative EQE of MQWs LEDs as a function of pulsed injection current density (0.1 % duty cycle and 1 kHz frequency).
Comparing the coupled-MQW-LB to the uncoupled-MQW-LB-LED, the peak EQE for
the former is 25% higher than that for the latter. We think that it is due to the relatively
enhanced hole transport through the thin and low InGaN barrier in MQW active regions
and resulting more efficient recombination of electrons and holes barring the
complication caused by the induced field. It is clear that the coupled MQWs with 3 nm
barriers would help provide a more uniform hole population among the 6-period quantum
wells than the uncoupled MQWs.44 In the uncoupled (12 nm barrier) MQW-LEDs, the
hole concentration dominates in the QW near the p-side, while for the coupled (3 nm
barrier) MQW-LEDs holes can be more uniformly distributed across all the QWs.
37
Therefore, all the 6 wells are more likely to participate nearly fully in the recombination
process in coupled MQW-LEDs, which thereby reduces the excess electron density and
thus electron leakage. As for the efficiency degradation ratio (taken as the EQE value at
600 A/cm2 relative to the maximum EQE), the LEDs with coupled MQW active layers
show smaller degradation ratios compared with the uncoupled MQW counterparts. A
coupled-MQW-LB-LED investigated shows negligible efficiency degradation while the
uncoupled-MQW-LB-LED counterpart shows 25% degradation, which can be attributed
to the enhanced hole distribution through thinner barriers.
3.2 Quantum efficiency for multi-DH LEDs
3.2.1 Motivation
As GaN-based LED technology continues to develop and mature, high brightness LEDs
retaining high quantum efficiencies at high injection levels (>100 A/cm2) have become
even more desirable to replace the prevailing incandescent lamps and fluorescent tubes in
general lighting. However, the quantum efficiency of typical InGaN multi quantum well
(MQW) LEDs peaks at current densities even as low as ~ 10 A/cm2, and drops with
increasing injection by a factor of as much as 2 in some reported cases. Although debates
still persist on the origins of “efficiency droop”, carrier overflow has been reported to be
the substantial component.45,46 In order to avoid carrier overflow and increase the light
output, LEDs must employ thick double heterostructure (DH) or MQW active regions.
In conventional InGaN/GaN MQWs, normally thick (>10 nm) GaN barriers were
employed to ensure active layer quality. However, electron and hole mainly recombine in
the QWs closest to p-GaN due to the poor hole transport.47,48 We have already exhibited
38
with employment of thin low height In0.06GaN barriers, hole population in the active layer
was improved, resulting higher quantum efficiency and suppressed efficiency droop at
high injection. DH active regions on the other hand can ensure more uniform hole
spreading across the active region due to the absence of barriers and consequently have
paved the way for negligible drop in quantum efficiencies beyond current densities of
~150 A/cm2.49 Moreover, DH LEDs possess bulk-like 3D density of states (DOS), and
therefore, can accommodate more carriers than thin QWs having constant 2D DOS.
However, among the ramifications of DHs are the degradation of InGaN structural
quality with increasing thickness and separation of electron and hole wavefunctions due
to the polarization field in the c-plane variety.50,51 Therefore, keeping the DH layer thin (3
nm) but stacking multiple of them separated by thin and low barriers (for ameliorating
hole transport) in the active regions could be a promising approach to maintain high
material quality and overcome the efficiency loss at high driving currents. In this work,
we demonstrate that by using 3nm-thick multi-DH active layers separated by 3 nm-thick
low-energy In0.06Ga0.94N barriers the electroluminescence (EL) is enhanced dramatically,
in proportion with the number of DH layers up to 4 without discernible efficiency loss at
high injection levels. Moreover, under resonant optical excitation, emission intensities at
10 K increase linearly with excitation power, indicating nearly unity quantum efficiency,
and scale with the active region thickness for a given excitation density. We have proved
that, with increasing number of active region DH layers, carrier overflow is
unequivocally shown to reduce significantly. Active layers quality degrades beyond 6 DH
layers due to the defects generation in the strain accumulation.
39
Among the ramifications of DH are the loss of InGaN quality and the increased band
bending due to the polarization field. In spite of this, we have demonstrated that with
increasing DH active layer thickness from 3 nm to 6 nm the relative EQE increased
considerably with a given SEI design. The relative EQE was enhanced substantially in
dual 3 nm DH and dual 6 nm DH LEDs, separated by a 3 nm-thick In0.06Ga0.94N barrier.
Incorporating more DH active regions of the same thickness, separated by thin and low
InGaN barriers, results in enhanced emission intensity without any discernible
degradation of the active region quality unlike that observed in thicker single DH layers
due to strain relaxation with increasing InGaN thickness.
3.2.2 Experimental procedures
The c-plane multi-DH InGaN LED structures, emitting at ~425 nm, were grown
on ~5 µm-thick n-type GaN templates on sapphire substrates in a vertical low-pressure
metalorganic chemical vapor deposition (MOCVD) system. The GaN templates
employed an in situ SiNx nanonetwork to reduce the dislocation density down to mid-108
cm-3.52 The active regions contained single or multiple In0.15Ga0.85N DH active layers
separated by 3 nm In0.06Ga0.94N barriers, shown in Table 3-1. All the structures
incorporate a staircase electron injector (SEI) for efficient thermalization of hot electrons
prior to injection into the active region and a 60-nm Si-doped (2×1018 cm-3) In0.01Ga0.99N
underlying layer for improving the quality of overgrown layers. The SEI consists of two
4 nm InGaN layers with step-increased In compositions of 4% and 8%, inserted in the
given order below the active region. The LED structures were completed with 100 nm-
40
thick Mg-doped p-GaN layers having 6×1017 cm-3 hole density, as determined by Hall
measurements on a separate calibration sample.
Table 3-1 LED structure of multiple DHs with sole SEI LED SEI Active layers
Single 4+4 nm 3 nm
Dual 4+4 nm 2x3 nm
Quad 4+4 nm 4x3 nm
Hexa 4+4 nm 6x3 nm
Octa 4+4 nm 8x3 nm
Single 6 nm 4+4 nm 6 nm
Dual 6 nm 4+4 nm 2x6 nm
Single 9 nm 4+4 nm 9 nm
Single 11 nm 4+4 nm 11 nm
41
3.2.3 Results and discussions
10-3
10-2
10-1
100
10-2
10-1
100
0 2 4 6 80
10
20
30
40
50
Single Dual Quad Hexa Octa
slop
e =
1
Inte
grat
ed P
L I
nten
sity
(a.u
)
Inte
grat
ed P
L I
nten
sity
(a.
u)
Excitation Power Density (kW/cm2)
(a) 10 K
slope =
1
(b) 295 K
IQE
(%
)
# of DHs
Figure 3.3: Integrated PL intensity as a function of excitation power density at (a) 10 K and (b) 295 K; gray solid lines indicate slope of 1 and the inset of (b) displays the PL-IQE vs. the number of 3 nm DHs in the active region; (c) PL efficiencies of multi-3 nm DHs vs. excitation power density at room temperature.
One common procedure to evaluate the IQE involves excitation-dependent
photoluminescence (PL) measurements and comparison of the PL intensities at low and
room temperatures by assuming 100% IQE at low temperature though it might not be the
case.53,54 Excitation power dependent resonant PL measurements were performed at both
10 K and 295 K using 385 nm excitation from a frequency doubled Ti:Sapphire laser
ensuring photo-generation of carriers only in the LED active regions. The highest
excitation density used corresponds to an average carrier concentration of ~1018 cm-3 in
the single DH LED structure. As the collected PL intensity is proportional to excitation
intensity, m
PL excL I , the linear dependence (m 1) for all structures at 10 K in Figure 3.3
(a) indicate that the radiative recombination dominates, i.e. Rad << nonRad, where Rad
and nonRad are the radiative and nonradiative lifetimes, respectively. It is therefore
42
reasonable to assume that the quantum efficiencies are nearly one at 10 K for all DH
LEDs, omitting the negligibly slight deviations. The room temperature data shown in
Figure 3.3 (b), however, shows superlinear dependence (m 1.4 - 1.95) for low
excitations, which is attributed to the notable impact of nonradiative recombination (m =
2 in case of constant nonRad). As the excitation density is increased, the slope gradually
approaches to m = 1 ( PL excI I ). The gradually decreasing slope in the intermediate
excitation regime indicates strong competition between nonradiative and radiative
processes and can be attributed to decreasing radiative lifetime, beneficial, for moderate
injections. Another process to be kept in mind is that with further increase of excitation,
saturation of localized states and delocalization of carriers (particularly holes, as the
electron density in the wells is in mid-1017 cm-3) may allow access to additional
nonradiative centers and result in enhanced recombination rate with respect to low
excitation. 55 , 56 Moreover, Coulomb screening of the quantum confined Stark effect
(QCSE) with increasing excitation leads to an increased interband recombination rate
1/Rad(N) BN, where B is the bimolecular recombination coefficient and N is injected
carrier density. The IQE values deduced from ratio of PL intensities at 300 K and 10 K at
the highest excitation density employed are shown in the inset of Figure 3.3 (b). The
quad 3 nm DH LED exhibits an IQE, so determined, of ~46% whereas increasing the
number of 3 nm DH active regions to 6 and 8 lowers the IQE to 36% and 16%
respectively, indicative of active region degradation with increasing overall thickness due
to plausibly strain relaxation and increased interface roughness. This degradation is
evident also from the room temperature PL efficiencies, defined as the collected
integrated PL intensity normalized to the incident laser power, shown in Error!
43
Reference source not found. (c). Notably, the PL efficiencies nearly scale with the
number of DH layers up to 6 due to increased absorption and emitting volume, showing ~
2, 4 and 6.5 fold increase for dual, quad and hexa DHs compared to single DH at an
excitation density of 1.5 kW/cm2, but no further improvement for the octa DH LED
which is most likely a manifestation of material degradation.
1 10 100 0 100 200 300 400 500
(b)(a)
Current density (A/cm2)
SingleDual Quad Hexa Octa
Inte
grat
ed E
L i
nten
sity
(a.
u)
slope =1
R
elat
ive
EQ
E (
a.u.
)
Figure 3.4: (a) The integrated EL intensity dependence on current density (the grey-sold line indicates slope of 1), (b) Relative EQE of multi-3 nm DHs vs. injected carrier density.
To study the impact of carrier overflow and other carrier transport features, we
measured EL efficiencies on-wafer (unpackaged) with light output collected primarily
normal to the sample surface by an optical fiber. The integrated EL intensities vs.
injection current are shown in Figure 3.4 (a). The integrated EL intensity, ELL , can be
described by a power dependence on the injection current density as mELL J , where the
power index m, as in the case of optical excitation, reflects an effective rate of
recombination processes within a given range of current densities. 57 The superlinear
44
growth of EL intensity (m ~ 1.4 for single, ~ 1.3 for dual and quad, and ~ 1.6 for hexa and
octa DH LEDs) at low current densities is attributed to nonradiative recombination.
Smaller m values suggest lower density of nonradiative recombination centers in single,
dual and quad DH LEDs compared to hexa and octa DH LEDs. The EL intensity changes
nearly linearly at high current levels; therefore, relative EQE tends to be constant.
As presented in Figure 3.4 (b), the relative EQE for the multi-DH structures with up to
4 DH layers increases rapidly with current injection and reaches its maximum at ~ 35-40
A/cm2. Compared to the single 3 nm DH LED, the peak EL efficiencies for dual and quad
DH LEDs are higher by 1.6 and 3.5 times, respectively. Unlike in the case of optical
injection, this significant improvement on EL efficiency cannot simply be explained by
increased emitting volume as for a given current density overall carrier concentration is
the same. Therefore, the data unequivocally indicates that increasing the number of 3 nm
DH layers (from 1 to 6) decreases the overflow of injected carriers considerably (i.e.
more of the injected carriers are captured by the active region), while further increase in
number of DH layers (8) aggravates by introducing more nonradiative recombination
centers due to degradation of the active region quality.
It is important to note that although the 4+4 nm SEI design used here has been shown
to be an effective replacement for the Al0.15Ga0.85N electron blocking layer for reducing
electron overflow in structures with multiple QWs and wider DHs (6 and 9 nm),58 it must
be optimized for a given active layer to fully prevent electron overflow. Based on the
simulation results in Figure 2.3, the percentage of electrons captured by and recombine
in the active region is increased to 76% in quad DH LED compared to 48% and 60% for
single and dual 3 nm DH LEDs, respectively, at a current density of ~500 A/cm2. LED
45
structures with optimized SEI layers should greatly reduce the carrier overflow if not
eliminated while maintaining the active region quality. For optimum SEI layer design,
which depends on the overall active region design, the resulting maximum EL
efficiencies for single and quad 3 nm DH LEDs should be similar.
As observed in Figure 3.4, the relative EQE of hexa 3 nm DH LED is only slightly
larger than that of the quad 3 nm DH, which suggests that the injected carriers are mostly
consumed in the first four DH layers close to p-GaN due to limited hole transport for the
achieved hole concentration and/or the active region quality may have slightly degraded
with increased overall thickness. Further increase in the number of DH layers to 8
lowered the EL efficiency by ~20%, indicative of a clear indication of the active layer
quality degradation, which is also confirmed by PL measurements conducted at 10 K and
295 K in Figure 3.3.
0 100 200 300 400 500 600
11 nm DH
Dual 3 nm DH Dual 6nm DH
3nm DH 6nm DH 9 nm DH
Re
lativ
e E
QE
(a.u
.)
Current density (A/cm2)
Figure 3.5: Relative EQE of DH LEDs as a function of pulsed injection current density (0.1 % duty cycle and 1 kHz frequency).
46
The relative EQE values for the LED structures with various DH thicknesses are shown
in Figure 3.5. Single 3 nm DH LED shows around 25% of peak EQE value compared
with that of the single 6 nm DH LED, indicative of severe electron overflow due to its
thin active region. Dual 3 nm DH LED shows doubled EQE value compared with single
3 nm DH LED. The single 6 nm DH LED structure shows the maximum relative EQE
values at current density ~41 A/cm2, slightly higher than that for 31 A/cm2 in 3 nm DH
LED, indicative of a slower rate of increase with injection for the 6 nm DH LED. The 9
nm DH LED shows a slow rise of EQE with increasing current injection and reaches its
maximum EQE at the injection current density as high as 220 A/cm2. The slow initial
EQE rise in the 9 nm DH is likely caused by both defects and slow increase of Beff
coefficient with injection current, which will be discussed later. The relatively low
maximum EQE can be ascribed to: (1) low value of radiative recombination coefficient
because of wide separation of electron and hole wavefunctions, and (2) possible onset of
strain relaxation giving rise to enhanced nonradiative recombination. Increase in DH
active region width to 11 nm results in the further decrease of the maximum EQE and
even slower EQE rise with injection current density. The 9 nm DH LED shows the
highest EQE among the single DH LEDs studied at injection current densities above
~180 A/cm2 with small efficiency degradation (~10% at 500 A/cm2 with respect to the
maximum). The dual 6 nm DH shows 20% higher peak EQE than that obtained from the
single 9 nm DH and much faster increasing rate of EQE with current injection. At 200
A/cm2, the relative EQEs retain 88% of the peak value for the dual 6 nm DH LEDs.
In conclusion, multiple DH structures have been demonstrated to enhance the quantum
efficiency of GaN-based LEDs at high injection levels. We showed that multi-3 nm DH
47
active layers (having 3D-like DOS) represent an effective avenue to improve quantum
efficiency with a given SEI design. Excitation dependent PL results indicate that PL
efficiency is nearly proportional to the number of multi-3 nm DH layers up to 6 at room
temperature, suggesting the same quantum efficiency for each DH active layer. Similarly,
EL efficiency is also shown to increase with the number of DH active layers up to 4, due
to reduced electron overflow, and the hexa DH LED shows ~20 % higher EL efficiency
than that of quad DH LED at high injection. Increasing single DH thickness from 3 nm to
6 nm resulted in an increase in peak EQE. Further increase of the DH active region
thickness to 9 nm improved EQE only at very high injection levels while 11 nm thick DH
showed significantly lower EQE due to active region quality degradation. Therefore,
among the efforts to enhance the quantum efficiency at elevated injection levels, multi-
DH layer designs with appropriate electron injectors can constitute a viable alternative
approach to achieve high efficiency and high power LEDs.
3.3 Optimization of stair-case electron injector
3.3.1 Motivation
We have undertaken a series of investigations to unveil the root cause of the efficiency
degradation mechanism. Experiments conducted on multi-DH LEDs on c-plane GaN
templates indicated that hot electron overflow is the dominant mechanisms in the EL
efficiency loss of LEDs at high current level. This clue shows that hot electrons, either
ballistic electrons or quasi-ballistic electrons, escape from the active region and
recombine with the holes in p-GaN layer where the recombination is predominantly
48
nonradiative. In this section, LEDs with various active regions have been designed. Some
of them incorporated single thicker active layer or multi-thin active layers with reduced
barrier height favoring hole transport. The structures were employed with varied
thickness SEI only (no EBL involved) to have a clear picture of SEI effects on electron
cooling in LEDs.
As discussed in last section, the EL efficiency of quad 3 nm DHs shows about 3.5 times
higher than that of single 3 nm DH. We attributed this to the more severe electron
overflow in single 3 nm DH. By incorporating more 3 nm DHs, the active layer thickness
is increased and thus electrons can be captured more readily in quad 3 nm DHs. First
order calculation also indicated that only 48% of electrons can be efficiently injected in to
the single 3 nm active region and recombine radiatively. As such, it is desirable to reduce
the electron overflow by optimizing SEI structures. Here, we increased the two-step SEI
thickness without modifying the In composition inside SEIs in order to cool down
electrons sufficiently by keeping the total SEI thickness up to 60 nm. Table 3-2 shows
the details of studied LED structures.
Table 3-2 LED structure with various active region designs and SEI thickness SEI thickness (nm) Active region (nm) LED-A 4+4 3 LED-B 20+20 3 LED-C 30+30 3 LED-D 4+4 Quad 3 LED-E 20+20 Quad 3 LED-F 30+30 Quad 3 LED-G 4+4 6 LED-H 20+20 6
3.3.2 Results and discussions
49
Before discussing the electron overflow in LED structures of different design, let us
evaluate the effect of SEI on the optical quality of the LED active regions. For this
purpose excitation density dependence of the resonant photoluminescence (PL) intensity
was measured at room temperature to gain insight into the optical quality of these DH
LEDs with varying SEI thickness, noting that in resonant excitation carriers are cool
already. A frequency-doubled Ti:Sapphire laser (385 nm wavelength) was used with the
maximum excitation density corresponding to an average carrier concentration of ~1018
cm-3 in the single DH LED. As shown in Figure 3.6 (a), PL intensity has a linear
dependence on the excitation power for all single and quad samples at 15 K, which
indicates that radiative recombination is dominant within the entire excitation density
range employed. 59 Moreover, the PL intensity scales nearly with the number of 3 nm DH
layers in the active region: PL intensity for quad DH LEDs is ~ 4 times that of single DH
LEDs. No change was observed in the emission wavelength with increasing SEI
thickness for any of the LEDs, suggesting that there is no discernible difference in strain
among the structures investigated. The PL intensities from the 6 nm DH LED structures
are between those for single 3 nm DH and quad 3 nm DH LEDs.
50
1E-3 0.01 0.1 1
(a)
single 3, 4+4 single 3, 20+20 single 3, 30+30 quad 3, 4+4 quad 4, 20+20 quad 3, 30+30 single 6, 20+20In
tegra
ted P
L in
tentie
s (a
.u.)
Excitation density (kW/cm2)
0.01 0.1 1
(b)
single 3, 4+4 single 3, 20+20 single 3, 30+30 quad 3, 4+4 quad 3, 20+20 quad 3, 30+30 single 6, 20+20In
tegra
ted P
L in
tensi
ties
(a.u
.)
Excitation density (kW/cm2)
Figure 3.6: The integrated PL intensity dependence on optical excitation density at (a) 15 K and (b) 295 K for single and quad 3 nm DH LEDs with varied SEI thickness.
Room temperature measurements [Figure 3.6 (b)] for the quad 3 nm DH LEDs
show that at low injection the slope of PL intensity dependence on excitation density
decreases with increasing SEI thickness, indicating reduced nonradiative recombination.
51
At a very low excitation density of 0.001 kW/cm2, increasing the SEI thickness from 4+4
nm to 20+20 nm improves the PL intensity by nearly an order of magnitude, while no
further improvement is apparent with the 30+30 nm-thick SEI. As the excitation density
is increased, the slope gradually approaches 1, indicating a strong competition between
nonradiative and radiative processes, with the latter becoming more prominent with
increasing injection. The improvement observed at low excitation densities with
increased SEI thickness suggests improvement of the active region material quality
manifested as a reduction of nonradiative recombination centers although particulars of
the genesis of this improvement is unclear at this instant. Similar trends are observed for
the single 3 nm DH LEDs at room temperature except lower PL intensities due to their
smaller active region thickness. Significant improvement is apparent in PL efficiency at
low excitation for the single 3 nm DH LED with increasing SEI thickness from 4+4 nm
to 20+20 nm. However, single 3 nm DH LED with 30+30 nm SEI exhibits the lowest PL
efficiency at low excitation densities but the highest intensity at the highest excitation
density employed. This is indicative of the presence of low density of certain type of
nonradiative centers that can be saturated faster with increasing excitation compared to
those in other LED structures. They may possibly originate from strain driven diffusion
of Mg from the p-GaN layer, the detrimental effect of which will be much weaker in the
quad LEDs, and/or generation of threading dislocations and related point defects within
the active region resulting from partial relaxation due to accumulation of strain energy
with increasing SEI thickness. Note that the detrimental effects of dislocations and point
defects acting as nonradiative centers can be stronger in the single DH LEDs, which is
consistent with the observations by Armstrong et al., 60 who reported reduction in density
52
of deep centers towards the LED surface, i.e. with increasing active region thickness,
within multi-QW or multi-DH LEDs. This unusual behavior for the single DH LED with
30+30 nm SEI is still under investigation; however, it does not affect the premise of this
paper as of particular importance is the LED operation at high excitation densities, where
the single 3 nm DH LEDs with 20+20 nm and 30+30 nm SEIs have comparable
performance. Moreover, it should be noted that the higher efficiencies (lower slopes) at
low excitation densities for the quad DH LEDs compared to the single DH LEDs result in
part from the improvement of material quality with further growth of InGaN active
regions in multi-DH LEDs. The integrated PL intensity from the single 6 nm DH LED
with 20+20 nm SEI exhibits the trend similar to those for the quad 3 nm DH LEDs with
20+20 nm and 30+30 nm, showing initially superlinear increase at low excitations and it
approaches slope one at high excitations.
To experimentally verify the overflow predictions, the integrated electroluminescence
(EL) intensity vs. current injection for single 3 nm and 6 nm DH LEDs and quad 3 nm
DH LEDs chips with varied SEI thickness was measured (0.1 % duty cycle and 1 kHz to
eliminate heating effects). Care was taken to assure invariance in the light collection
geometry among all the chips tested. Moreover, the light emission intensity showed less
than 5% variation among at least five devices measured on the same wafer.
53
10 100sl
ope
=1
(a)
single 3, 4+4 single 3, 20+20 single 3, 30+30 quad 3, 4+4 quad 3, 20+20 quad 3, 30+30 single 6, 4+4 single 6, 20+20In
tegra
ted E
L Inte
nsi
ties
(a.u
.)
Current density (A/cm2)
0 100 200 300 400 500
(b) quad 3, 4+4 quad 3, 20+20 quad 3, 30+30 single 6, 20+20
Rela
tive E
QE
(a.u
.)
Current density (A/cm2)
single 4, 4+4 single 4, 20+20 single 4, 30+30 single 6, 4+4
Figure 3.7: (a) The integrated EL intensity dependence on current density for DH LEDs with varied SEI thickness. (b) The relative EQE vs. injected current density.
Figure 3.7 (a) demonstrates that for the LEDs with 4+4 nm SEI, the EL efficiency of
quad 3 nm DH LED is ~3.5 times higher than that of single 3 nm DH LED. It is obvious
that the low EQE of the single 3 nm DH LED with 4+4 nm-thick SEI is due to higher
54
carrier overflow in the thinner active layer facilitating ballistic and/or quasi-ballistic
electron transport across the active layer without recombination. This is in good
agreement with the calculated electron overflow percentiles shown in Figure 2.3, which
predicts substantially higher overflow for the single 3 nm DH LED.Error! Bookmark
not defined. The SEI thickness should be larger for thin active region LED like single 3
nm DH to provide hot electrons more time to sufficiently thermalize before being
injected into the active region. However, regardless of the active layer design, the total
SEI thickness must be kept below the critical thickness above which the active region
material quality would degrade noticeably due to strain relaxation. It should also be noted
that the agreement between the simulation results of Figure 2.3 and the data presented in
Figure 3.7 is satisfactory considering that the calculations assume a constant phonon
scattering rate, which would result in underestimation of the overflow percentiles as the
phonon lifetime has been reported to decrease with increasing current injection.61
Increase of the SEI thickness from 4+4 nm to 20+20 nm for the single 3 nm DH LED
resulted in an enhancement in the peak EL efficiency by nearly 3.5 times, making it
comparable to those of the quad 3 nm DH LEDs [Figure 3.7 (a)] and the single 6 nm DH
with 4+4 nm SEI. This significant improvement indicates that the electrically injected hot
electrons are cooled more efficiently within the thicker SEI. The more visible efficiency
roll off in the single 3 nm DH LED both with 20+20 and 30+30 nm-thick SEIs compared
to the single 6 nm and the quad 3 nm DH LEDs [Figure 3.7 (b)] suggests larger electron
overflow in the thinner active region, which increases at higher injection levels. At the
highest injection level employed (550 A/cm2, which corresponds to an average carrier
density around 5 x 1018 cm-3, estimated using A = 107 s-1 and B = 10-10 cm-3s-1)62, the
55
efficiency for the single 3 nm DH LED with 20+20 nm SEI or 30+30 nm SEI is nearly
twice that for the single DH LED with 4+4 nm SEI and 55% to 70% of those for single 6
nm and quad 3 nm DH LEDs. It should be noted that thicker, multi-DH or MQW active
regions would act as electron coolers naturally. Accordingly, the single 6 nm DH and the
quad 3 nm DH LEDs with 20+20 nm SEI show only a moderate increase of 15-20% in
the peak EQE compared to those for the LEDs with 4+4 nm SEIs. At injection current
densities below 200 A/cm2 (the range of the practical interest for LED applications), the
quad 3 nm DH and the single 6 nm DH with 20+20 nm SEI show higher relative EQE
values due to the reduced overflow as compared to their counterparts with 4+4 nm SEIs
and retain about 87% and 79% of their peak EQEs at 150 and 200 A/cm2, respectively.
In conclusion, by optimizing the SEI structures, mainly increasing the thickness, the EL
efficiency has been improved considerably for single 3 nm DH with 20 + 20 nm or 30+30
nm SEI owing to the reduced electron overflow. However, further increasing the SEI
thickness degraded the material quality which offers lower EL efficiency and PL intensity
as well. Further optimization for this structure including slight reducing the total
thickness needs to be done.
3.4 Delta p-doped barrier in LED active regions
3.4.1 Motivation
Among the mechanisms responsible for the efficiency degradation at high injections,
one obvious cause of the efficiency roll-over is the asymmetrical doping in GaN,
resulting in an insufficient supply of holes to the active region compared to electrons
concentration under electrical injection. Because hole injection lags behind electron
injection, the radiative recombination cannot keep up with increasing carrier injection.
56
These results in either electron escape to p-GaN without recombination or electron
accumulation in the active region, which reduces radiative recombination efficiency. To
address this problem, the efforts were focused on either increasing hole concentration in
p-GaN or doping barriers in the multi-active region LED structures.30,63,64,65 As regarding
to the second approach, because active regions close to n-GaN are lack of holes, carriers
recombine mainly in the DH adjacent to p-GaN due to the poor hole transport.66 ,67
However, Mg doping in the active region has a detrimental effect on LED efficiency,
since Mg acts as luminance “killer” that results in low quantum efficiency of LEDs with
doped active regions.68 So far, there is no experimental report incorporating Mg into the
active region without sacrificing the LED efficiency.
Hence, we proposed the employment of p-type δ-doping barriers, i.e. Mg was
solely doped into barrier during interruption of barrier growth, to improve the hole
injection for higher quantum efficiency from one aspect and to eliminate or at least to
mitigate possible Mg out-diffusion from barriers to the InGaN active region from another
aspect. The experimental data are augmented with numerical simulation of electron and
hole concentration in the active regions of LED structures.
3.4.2 Experimental procedures
The c-plane InGaN LED structures emitting at ~430 nm were grown on ~4 µm-
thick n-type GaN templates on sapphire in a vertical MOCVD system. The LED
structures, dubbed in the following discussion as samples A, B and C, are shown in
Figure 3.8 (a), (b) and (c), respectively.
57
Figure 3.8: The schematics of hex 3 nm DH LED structures for the improvement of hole injections. Two 6 nm InGaN quantum barriers are located at the n-side for possible p-doping, while the other barriers are 3 nm. (a) In LED A, all the barriers are kept undoped. (b) In LED B, the first 6 nm barrier closest to n-side is p-doped. (c) In LED C, the two 6 nm barriers close to n-GaN were p-doped.
All three LEDs are featured by hex (6x) 3-nm In0.15Ga0.85N DH active regions separated
by relatively low energy In0.06Ga0.94N barriers to enhance the hole transport across the
active regions. The three In0.06Ga0.94N barriers near p-GaN are 3 nm-thick and the two
In0.06Ga0.94N barriers near n-GaN are designed to be 6 nm-thick to prevent possible Mg
out-diffusion from p-doped barriers into the active regions. The δ-doping with Mg was
implemented as follows: after the first 3 nm growth of the 6 nm-thick barrier, the growth
was interrupted, and the Cp2Mg source flew into the growth chamber for 20 s, followed
by the growth of remaining 3 nm of the 6 nm barrier. In the reference structure (sample
A), all the barriers are kept undoped. In LED B, only the 6 nm barrier closest to n-side is
58
p-doped, while others parts are undoped. In sample C, both 6 nm barriers are p-doped.
The hole concentration in the doped barrier was estimated to be 4×1017 cm-3, based on
delta-doped p-GaN deposited in separate experiments. A 15+15 nm-thick SEI consisting
of In0.04GaN and In0.08GaN layers is situated beneath the active region on the n-side to
efficiently cool injected hot electrons. 69,70 A 60 nm Si-doped (2×1018 cm-3) In0.01Ga0.99N
underlying layer was inserted between SEI and n-GaN for quality improvement. The
LED structures were completed with 100 nm-thick Mg-doped p-GaN layers with 6×1017
cm-3 hole density, as determined from Hall measurements performed on separate samples.
Square mesa patterns (400×400 μm2) were formed by conventional photolithography and
chlorine based Inductively Coupled Plasma (ICP) etching. Ti/Al/Ni/Au (30/100/40/50 nm)
metallization annealed at 860 ºC for 60 seconds was used for n-type Ohmic contacts, and
5-nm/5-nm Ni/Au film served as the semi-transparent p-contacts with 40/50-nm Ni/Au
contact pads completed the LED fabrication.
3.4.3 Numerical Simulation of delta p-doped barrier LEDs
In order to verify the efficiency of delta Mg-doped barriers in improvement of hole
injections for LEDs, electron and hole distributions were simulated for the active regions
of the investigated LED structures under an injected current density of 100 A/cm2 using a
SILVACO ATLAS commercial simulator with simulation parameters appropriate for
nitride materials (A = 107 s-1, B = 10-10 cm-3s-1, and 40% of the theoretical polarization
charge was assumed) in Ref. 32.
59
Figure 3.9: (a) Electron concentration, (b) hole concentration, and (c) band structure) simulated for sample A (black), sample B (red) and sample C (blue) under current injection 100 A/cm2 with SILVACO ATLAS with parameters appropriate for nitride materials. Figure 3.9 (a) exhibits distribution of electrons in the active regions of the three LED
structures under investigation for an injection current density of 100 A/cm2. As evidenced
from the figure, due to efficient cooling of injected electrons in the SEI, the electron
60
distribution is relatively uniform in sample A. Electron concentrations in the first DH
near n-GaN are quite similar for all three LEDs, while the second DH in sample B and C
exhibits lower electron density compared with sample A. This effect is likely caused by
reversed polarization field induced by p-type doping of In0.06Ga0.94N barrier, which could
also be seen in Figure 3.9 (c). This could also explain the difference in electron
concentrations in the third DH, where the electron density is lower in sample C compared
to those in sample A and B.
Usually hole transport deep into multilayer active regions of the InGaN-based LEDs is
impeded owing to their relatively large effective mass and low mobility. Therefore,
without considering hot electron overflow, radiative carrier recombination is limited by
the concentration of holes since holes are the minority carriers in the active region under
bias. Under these circumstances, as shown in Figure 3.9 (b) for the reference sample A,
the hole distribution across the active region is very non-uniform, with the highest
concentration in the DH closest to p-GaN and a rapid decrease in hole density towards
the n-side of the LED structure. Therefore, for the conventional LEDs, electron and hole
recombination occurs mainly in the DH adjacent to p-GaN, while the rest of DHs are lack
of holes for recombination. When the InGaN barrier closest to the n-GaN is δ-doped with
Mg (sample B), the hole concentration in the nearby DH active regions (the first and the
second DH) increases significantly, greatly reducing the asymmetry of electron and hole
injection and thus enhancing the recombination efficacy. When two 6 nm-thick InGaN
barriers are δ-doped (sample C), the hole concentration in DHs near n-GaN was further
increased, particularly in the DH second closest to the n-GaN side, which would further
improve the LED output power.
61
Figure 3.9 (c) shows the simulation results of the band structure for the three LED
structure in the active region under current density 100 A/cm2. The δ-doping of the
In0.06GaN barrier closest n-GaN increases the effective height of the barrier in the
conduction band (barriers height to quasi-Fermi level is 154 meV, 190 meV, and 192
meV in sample A, B and C, respectively). As a result, the confinement of electrons in the
first DH near the n-GaN is improved with δ-doping of the In0.06GaN barrier.
3.4.4 Results and discussion
To demonstrate the effect of p-type delta-doping of barriers on LED performance,
integrated EL intensities and relative external quantum efficiency (EQE) were measured
for sample A, B, and C under pulsed excitation with 0.1 % duty cycle to eliminate heating
effects. As shown in Figure 3.10 (a), sample B and C exhibit higher integrated EL
intensities for high current injection levels (> 50 A/cm2), indicating no obvious material
degradation associated with Mg incorporation into the active regions compared to the
reference structure. As shown in Figure 3.10 (b), the peak value of relative EQE for
sample B was ~20% higher than that for sample A. The δ-doping of two barriers in
sample does not increase the EQE peak value further compared with sample B, which
possibly indicates cancelling the positive effect from further improvement of hole
injection by reduced active region quality due to the presence of higher amount of Mg in
the active region. As an additional benefit of δ-doping, samples B and C suffer less
efficiency reduction at high injection compared to sample A (23% for sample A vs. 14%
for sample B and C at 400 A/cm-2). This finding implies that the improved hole injection
partly suppress the efficiency droop at high injection level. EQE peak roll-off in LEDs
62
with delta-doped barriers shifts towards higher current (80 A/cm-2 for sample A, 120
A/cm-2 for sample B and C).
0 100 200 300 400 0 100 200 300 400
Current density (A/cm2)
Re
lative
EQ
E (
a.u
.)
Inte
gra
ted E
L in
ten
sitie
s (
a.u
.)
Sample A Sample B Sample C
Figure 3.10: (a) The integrated EL intensities and (b) the relative EQE vs. injected current density for sample A, B, and C.
In conclusion, we have shown that δ-doping with Mg of barriers significantly improves
the quantum efficiency of LEDs with active regions composed of hex 3-nm DHs. With
careful design of the active regions, the improvement in hole injection significantly out-
weights the possible degradation of the material quality due to the presence of Mg. With
Mg δ-doping of the first barrier on the LED n-side, the relative peak EQE improves by
20% in compared with the reference structure. Doping of the first and second barriers on
the n-side does not result in further improvement of relative EQE compared to the case of
one doped barrier. This possibly indicates that further improvement in hole injection is
canceled by the reduced active region quality due to higher Mg-doping. EQE peak roll-
off in LEDs with δ-doped barriers shifts to higher current indicating the reduction of
63
asymmetry in injection of electrons and holes with the employment of δ-doping. The both
LED structures with Mg delta-doped barriers effectively suppress the efficiency droop at
high injection level compared to the counterpart with un-doped barriers.
3.5 Investigation of GEI LEDs
We examined LED structures with different active-region and SEI designs and aimed at
reducing the electron overflow (better efficiency retention at high injection current
density) as well as improving the quantum efficiencies. In Chapter 2, we already
theoretically investigated GEI and SEI on the effects of electron cooling, and GEI
exhibits better performance in suppression of electron overflow than SEI with the same
total electron injector thickness. In this realm, 6x2 nm and 6x1.5 nm MQWs with GEI: 20
nm were analyzed and compared with the best LEDs with two-step SEI designs.
0.0 0.5 1.0 1.50
20
40
60
80
(a)
IQE
(%
)
Excitation density (kW/cm2)
quad 3 nm, 20+20 nm SEI single 6 nm, 20+20 nm SEI hexa 2 nm, 20 nm GEI hexa 1.5 nm, 20 nm GEI
64
0 100 200 300
(b)
Rela
tive E
QE
Current density (A/cm2)
quad 3 nm, 20+20 nm SEI single 6 nm, 20+20 nm SEI hexa 2 nm, 20nm GEI hexa1.5 nm, 20nm GEI
Figure 3.11: (a) IQE of LEDs determined from the ratio of room temperature to low temperature PL intensities at the maximum excitation density employed, assuming unity internal quantum efficiency at 15 K. (b)The relative EQE vs. injected current density. We can see that, in Figure 3.11 (b), with the same active region thickness, hexa 2 nm
MQW LED shows 20% higher EL efficiency than quad 3 nm DH LED even MQW LED
was employed with thinner electron injector (20 nm versus 40 nm). If compared with
single 6 nm with 20+20 nm SEI, hexa 2 nm MQW shows 10% higher peak EQE value
and much less efficiency droop at high injection up to 300Acm-2. Moreover, hexa 1.5 nm
MQW LED shows nearly double EQE peak value compared with that of quad 3nm DH
LED. The mechanism of this great EL intensity improvement is still under investigation.
65
3.6 Injection-dependent radiative recombination coefficient (B) of
single and multi active layer DH LEDs
The radiative recombination coefficient, B, is dependent on injection through a variety
of processes. Change in junction temperature due to Joule losses is one such process as B
coefficient is inherently temperature dependent. Carrier lifetime, which is related to the B
coefficient, increases with temperature which implies that the B coefficient reduces with
increasing junction temperature exacerbated by limitations in heat transfer.71 The B
coefficient is also dependent on injection due to the skewed band edges in the active
region of LEDs and also on the total number of states that are available. In biaxially
strained InGaN/GaN layers, the mismatch strain induces a polarization field (allowed by
the wurtzite-structure lattice symmetry) along the growth direction72. This strain induced
piezoelectric field and spontaneous polarization field tilts the potential profile, resulting
in triangular quantum wells in the hetero-interfaces of active regions. Consequently, the
optical transition energy is reduced. Therefore, with increasing current injection, a
blueshift will occur and the magnitude of the blueshift depends on the active layer
thickness 73 . The simulated potential profiles, using the commercial Silvaco ATLAS
software package, are shown in Figure 3.12 (a) and (b) for single 3 nm and the 9 nm DH
LEDs, respectively, at different injection current densities. Note that a factor of 0.4 was
used for the polarization induced charge for both electrons and holes at the interfaces in
generating Figure 3.12 (a) and (b). In a thick active layer like single 9 nm DH, electron
and hole wavefunctions are widely separated at low injection levels, which results in
reduced radiative recombination rate and relatively low EQE. We have to note that the
available states of the triangular potential well are not completely filled at low current
66
injection. With increasing the injection level, the triangular region will be fully occupied
due to relatively low density of states. With further increase in injection, the quasi-
continuum states followed by the three dimensional states in the DH active region will
begin to fill, leading to higher recombination rate and thus higher relative EQE.
0 10 20
-3
0
1
0 10 20
500, 50, 0 A/cm2
0, 50, 500 A/cm2
0, 50, 500 A/cm2
0, 50, 500 A/cm2
(a)
Ener
gy (
eV)
0, 50, 500 A/cm2
500, 50, 0 A/cm2
0, 50, 500 A/cm2
0, 50, 500 A/cm2(b)
Distance (nm)
67
0 200 4000
10
20
30
40
50
60
70
80
(d)
Sim
ulat
ed E
nerg
y sh
ift
(meV
)3nm6nm9nm11nmDual 3nmDual 6nm
Exp
erim
enta
l E
nerg
y sh
ift
(meV
)
Current density (A/cm2)
(c)
0 200 400
Figure 3.12: Energy band edge profiles simulated for (a) 3 nm DH and (b) 9 nm DH LEDs with 4+4 nm SEIs at different injection current densities. Peak emission energy shift as a function of injection current density from (c) EL measurements and (d) Silvaco simulations for DH LED structures with various active regions. Figure 3.12 (c) shows the observed peak emission energy shift as a function of injection
current density from electroluminescence (EL) measurements. Upon current injection, the
polarization fields in the strained InGaN DH layer are screened by free carriers, which
weakens the quantum confined Stark effect (QCSE), and then increases the transition
energy resulting in the blueshift in emission. Larger peak energy shifts were observed
with injection when the DH thickness is increased from 3 nm to 9 nm, as presented in
Figure 3.12 (c). The entire DH LEDs exhibit a large portion of the energy shift at low
injection levels (< 50 A/cm2) followed by a slower shift with further increase in injection.
This is attributed to the presence of triangular potential wells at the interfaces, which are
filled rather rapidly even at relatively low injection levels causing the large initial
blueshift. Recombination from the higher lying quasi-continuum states and 3D states in
68
the DH active region start to contribute to emission with increased injection, resulting in
the slower blueshift observed.
For single 6 nm, dual 6 nm, and single 9 nm DH LEDs, a significant 25 meV energy
shift was observed with initial increase of current density to 40 A/cm2, while only a 8
meV shift is evident for the 3 nm DH LED within the same current density range. Figure
3.12 (d) shows the simulated peak energy shift as a function of current density obtained
using Silvaco ATLAS. The polarization charge densities of 7.7x1012 cm-2 and
3.8x1012cm-2 have been used for the interfaces between the In0.01Ga0.99N barrier and the
In0.15Ga0.85N active layer, and the In0.15Ga0.85N active layer and the In0.08Ga0.92N SEI,
respectively, which are within the range of reported values74, and provide energy shifts in
the range comparable to those observed experimentally. The tendency is consistent with
the experimental results except for the single 9 nm DH and 11 nm DH structure. This
observed discrepancy is likely due to the strain-driven material quality degradation for
thick single DH active region (9 nm and 11 nm DH) neglected in simulations. We should
also note that varying the polarization charge in the simulations changes the absolute
energy shift, but does not affect the peak energy shift dependence on the active layer
thickness.
We now discuss the electron and hole wavefunction overlap in the context of
radiative recombination Beff coefficient as a function of injection current density in a
system with confinement along the z-direction (growth direction). Following the
formalism of Ref. 75, in the realm of Fermi’s Golden Rule, the spontaneous transition
rate from a group of initial states i in the conduction band to a group of final states f in
the valence band separated by a transition energy hω can be expressed as
69
:
2
22 *0
0
2
2i f f i r
eAT M z z dz F
m
(3-1) where A0 is the magnitude of the sinusoidal local vector potential, e is the electron charge,
m0 is the free electron mass, M is the in-plane momentum matrix element, r is the
reduced density of states, is the transition energy, 1c vF f f is the Fermi factor
given in terms of the Fermi functions for the conduction (fc) and valence bands (fv),
xyz r r are the envelope functions representing the wavefunctions. As noted
in Eq 3-1, the spatial overlap between the electron and hole wavefunctions ( e and h )
is obviously imperative, and the radiative recombination rate is proportional to the
squared overlap integral when electrons and holes are confined in the z-direction..
For quantum-confined structures, it has been suggested that low-dimensional
equivalents of the bimolecular radiative recombination B coefficient should be introduced
to eliminate the artificial dependence of the radiative recombination current on size, such
as the active region width in two-dimensional (2D) systems.75 By defining the
spontaneous transition rate as 2 2 2spont D D DT B n p where 2Dn and 2Dp are the 2D electron
and hole densities, respectively, the 2D B coefficient in InGaN quantum wells with
confinement along the z-direction can be expressed in terms of the momentum matrix
element of Eq 3-2:
70
222
2 *2 3 * *
0 0 0
4
2D h e r
B e h
n eB M z z dz
c k T m m m
,
(3-2)
where n is the refractive index, 0 is the permittivity of free space, c is the speed of light,
is photon energy, kBT is the thermal energy, and *em and *
hm are the electron and hole
effective masses. The momentum matrix element M can be obtained from the in-plane
interband transition matrix element (for polarization within the plane), 2cvP M , 76
which has been determined from the absorption measurements for binary constituents
InN and GaN.77 Using Pcv = 9.6x10-20 g cm/s obtained from linear interpolation for the
required composition, we calculated the B2D coefficient to be 1.8x10-4 cm2s-1 for an
In0.15GaN active region assuming full overlap of electron and hole wavefunctions. For 3D
rate equation, applicable to 3D DH structures, the 2D B coefficient should be multiplied
by the active region thickness, zL . To test this approach and the validity of the 2D
approximation, the 3D limit for the B coefficient for In0.15Ga0.85N was also calculated
from the following equation78
3/22 2
23 1 22 3 2
0
2 1D
B x y z
e nB M
m c k T m m m
,
(3-3)
where , , , , , ,x y z e x y z h x y zm m m . The value of 3D B coefficient derived from Eq. 3-3 is
5x10-11 cm3s-1 for In0.15Ga0.85N. This value is smaller than that obtained using B2DLz
even for the thinnest active region with Lz = 3nm investigated in this study. We,
therefore, assume that all LEDs with active region widths of 3 nm and above exhibit 3D
behavior but with an electric field along the growth direction reducing the spatial
71
overlap of charge carrier distributions in the active region. Therefore, the injection
dependent overlap integral of the electron and hole wavefunctions should be
incorporated into the calculation of the 3D Beff coefficients using 5x10-11 cm3/s as the
upper limit for full overlap:
2
11 3 -1 *
0
5 10 cm seff h eB z z dz
(3-4)
0 100 200 300 400 500
0.0
0.5
1.0
1.5
2.0
2.5
3.0
0 100 200 300 400 500
(a)
Bef
f coe
ffic
ien
t (1
0-1
1 cm
3/s
)
3nm, SEI 4+4 3nm, SEI 20+20 6nm, SEI 4+4 6nm, SEI 20+20 9nm, SEI 4+4 9nm, SEI 20+20
Current density (A/cm2)
(b)
4x3nm, SEI 4+4 4x3nm, SEI 20+20 2x6nm, SEI 4+4 2x6nm, SEI 20+20
Figure 3.13: Calculated coefficients Beff of (a) single DH, and (b) multi DH LEDs, calculated using squared overlap integrals of electron and hole wavefunctions (proportional to radiative recombination rate) within the active region as a function of current density using SILVACO ATLAS software package. The SEI layer thicknesses are provided in the legends in nm units.
72
Figure 3.13 represents the simulated bimolecular recombination coefficients, Beff,
derived from the transition matrix element and the simulated squared overlap integrals of
the electron and hole wavefunctions in the single and multi DH active regions. It is
apparent from Figure 3.13 that the radiative recombination coefficient, instead of being
constant as assumed,79 in fact depends on injection and particulars of the active-region.80
The Beff coefficient initially increases with injection and then tends to saturate at high
injection levels as the flat band condition is approached. Thinner active layers (single 3
nm DH) exhibit larger Beff values at low injections, which is attributed to their relatively
larger spatial overlap of the electron and hole wavefunctions. At ~200 A/cm2, the single
3 nm DH LEDs exhibit about 37% and 60% higher squared overlap integral value
compared to the single 6 nm and the single 9 nm DH LEDs, respectively. The change in
the Beff coefficient with injection is very small for the single 3 nm DH LEDs, while that
for wider DHs exhibits substantial increase of Beff coefficient within injection in the range
of 0 A/cm2 to 500 A/cm2. The smaller Beff coefficients in wider active regions are,
naturally, attributable to increased spatial separation of electrons and holes caused by the
polarization fields.81
It is worth noting that wider SEIs give rise to larger overlap integral for otherwise the
same active layer thickness. The difference in the overlap integrals for the single and the
dual 6 nm DH LEDs with 4+4 nm and 20+20 nm SEIs exhibiting 18% and 22% larger
Beff coefficients, respectively, at 100 A/cm2. When injection approaches 500 A/cm2, this
difference is somewhat reduced to 13% and 20% for the LEDs with 4+4 nm and 20+20
nm SEIs.
73
It has been reported that radiative and nonradiative coefficient is proportional to square
of electron and hole wave functions overlap.82 In order to get the dependence of radiative
and nonradiative coefficient on temperature, we can use the radiative and nonradiative
decay time to extract them. Our assumption is that radiative process is dominant at low
temperatures, i.e. the internal quantum efficiency is 100% at low temperature. The PL
intensity is proportional to the ratio of nonradiative ( nr ) decay times to total decay times
as following:
1I(T)
1
nr
rr nr
nr
(3-5) where I(T) is PL intensity as a function of temperature. If we normalize the PL intensity
to unity at 10 K, then we can obtain the following equation:
1I(T) I(10K)
1 r
nr
(3-6) where I(10K) is the PL intensity at 10 K. This quenching behavior which is typical for
nitride semiconductors is caused by thermal activation of nonradiative recombination
channels and/or localization energy above a certain temperature.
Note that the measured PL decay time can also be expressed in terms of radiative ( r )
and nonradiative decay times ( nr ) as
74
1 1 1
PL r nr
(3-7)
where (T)PL is the total or effective PL decay time and also the quantity obtained
from fitting the PL transients.
Using the Eq. 3-6 and 3-7, one can reach the temperature dependence of the radiative
and nonradiative decay times as a function of temperature.
0 100 200 30010
0
101
102
103
104
0 100 200 3000 100 200 300
PL
R
Nr
Dec
ay T
ime
[ns]
(a) (c)
PL
R
Nr
Temperature [K]
(b)
PL
R
Nr
Figure 3.14: The PL decay time ( PL ), radiative decay ( R ) time and nonradiative
decay time ( NR ) for the hexa 1.5 nm (a), 2 nm (b) and 3 nm (c) LEDs as a function
of temperature The results are shown in Figure 3.14. As can be seen, radiative recombination is
dominant process at low temperature. Up to 100 K radiative decay time is almost
independent of temperature, indicative that radiative decay time is determined by the
contribution of bound excitons. Nonradiative recombination rate is very low below 100
K, because bound excitions will mainly have radiative recombination at low temperature,
75
while nonradiative recombination centers (deep level defects in the layer) have negligible
effect at low temperature. Only a decrease of the emission intensity can be expected with
increasing temperature because of ionization of the bound excitons at this temperature
range. The carriers are thermally activated and reached defects where they can recombine
nonradiatively above 100 K for all samples. The radiative to nonradiative transition point
at which the nonradiative recombination rate starts to exceed radiative recombination rate,
appeared at almost same temperature for all LEDs. With increasing temperature, bound
excitions break out and free electrons and holes recombination dominant in carriers
recombination. Therefore, nonradiative lifetime decreases with increasing temperature.
The reason for the increase in radiative lifetimes with increasing temperature could be
attributed to the decreasing probability of overlapping of electron and hole wave function
because of the screening of the phonon emission.
Using the Eq. 3-8 and 3-9, we can reach the temperature dependence of the nonradiative
and radiative coefficient.
1
nr
A
(3-8) 1
r
Bn
(3-9) It is clear that A coefficient increases with increasing temperature and nonradiative
recombination is dominant in room temperature. In contrast, B coefficient decreases with
increasing temperature and radiative recombination is dominant in low temperature.
76
3.7 Investigation of layer quality on semi-polar GaN
3.7.1 Motivation
As already mentioned earlier, in biaxially strained InGaN/GaN layers, the mismatch
strain induces a polarization field (allowed by the wurtzite-structure lattice symmetry)
along the growth direction. This strain induced piezoelectric field and spontaneous
polarization field tilts the potential profile, resulting in triangular quantum wells in the
hetero-interfaces of active regions. Especially in a thick active layer, electron and hole
wavefunctions are widely separated at low injection levels, which results in reduced
radiative recombination rate and relatively low EQE. Therefore, nonpolar and semipolar
GaN substrates, where the spontaneous polarization field is zero or much weaker than
polar c-GaN, have prospective use in optoelectronic devices due to larger wavefunction
overlap of electrons and the holes in active regions increasing the quantum efficiency.
3.7.2 Experimental procedure
To achieve (1101) GaN film surface parallel to the substrate, Si (001) substrates that are
miscut 7o toward the Si 110 direction were used. The substrates were patterned to form
grooves aligned parallel to the Si 110 direction as described elsewhere.83,84 The top
terraces were 3 um wide, while grooves were either 10 or 3 um wide (hereafter, 3 u m x
10 um and 3 um x 3 um patterns, respectively). A 50-nm-thick AlN layer was grown by
MOCVD to serve as a seed layer for GaN growth. SiO2 mask layer was formed on one of
the sidewalls of the groove, and subsequent growth of GaN started on the opposite side of
the groove. The substrates were then reloaded into the MOCVD chamber and GaN
77
growth was performed with the use of trimethylgallium (TMG) and NH3 as sources of Ga
and N, respectively. Semi-polar GaN layers were grown at a chamber pressure of 200
Torr and high NH3 flow rates (>2000 sccm), which were found to be favorable for the
(1101) facet formation. The growth initiated on Si {111} sidewalls free from SiO2 and
then advanced laterally first along the GaN [0001] c+ direction and then additionally
along the [0001] c- direction after the vertical growth advanced above the Si (001)
terraces, which are shown in Figure 3.14.
Figure 3.15: (a) Cross-sectional SEM image of coalesced semi-polar GaN layers with 3 um x 3 um pattern; (b) inclined SEM image of non-coalesced semi-polar GaN layers with 3 um x 10 um pattern.
3.7.3 Results and discussion
A series of optical measurements were performed to investigate the quality of semipolar
(1101) GaN grown on patterned Si (001) substrates. A c-plane (0001) GaN/sapphire and
a c-GaN layer grown on the state-of-art GaN template using in-situ epitaxial lateral
overgrowth with SiNx nano-network to block dislocation propagation (referred as nano-
78
ELO layer) were measured as reference samples. Figure 3.15 compares the room
temperature PL spectra from the c-plane GaN film on sapphire, nano-ELO layer, non-
coalesced and coalesced semipolar (1101) GaN layers. We can see that the intensity of
near-band-edge (NBE) from non-coalesced (1101) GaN is comparable to that of the c-
plane layer grown on sapphire. The PL intensity of coalesced GaN layer is much weaker,
which can be due to defects formed in the layer as a result of coalescence and relatively
larger total area of the defective c– wings.85
Figure 3.16: Steady-state room-temperature PL spectra of c-plane GaN films grown on sapphire and (1 100) m-plane and (1101)-oriented GaN layers grown on the Si patterned substrates.
Normalized TRPL data were fit using a biexponential decay
function -t / τ -t / τ1 2A e + A e1 2 , where A1 and A2 are the amplitudes of the slow and fast
decay components with representative time constant τ1 and τ2, respectively. Figure 3.16
79
compares room-temperature excitation density dependent TRPL for c-plane nano-ELO
GaN on sapphire and non-coalesced (1101) GaN on Si (001), and Table 3-3 lists the
fitting parameters for c-plane and (1101) GaN layers. The normalized TRPL data were fit
by using the bi-exponential decay function. As seen from Figure 3.16 and Table 3-3, the
slow decay component τ2 for both layers becomes longer with increasing excitation
power density. It indicates that nonradiative and radiative recombinations compete at low
excitations, while radiative recombination becomes dominant at high excitation density.
The decrease of radiative recombination time with increasing excitation power density is
attributed to increased electron and hole overlap resulting in faster radiative
recombination. The slow decay component for the non-coalesced layer is as long as 1.98
ns, which is even longer than that of the state-of-the-art nano-ELO GaN layer (1.15 ns). 86
This long slow decay constant is indicative of high optical quality of the noncoalesced
layer and lower density of extended and point defects compared to coalesced layer. In the
case of semipolar GaN, regions of high quality c+ wing regions are probably responsible
for the long slow decay components τ2 at high excitation levels.
Figure 3.17: Excitation density dependent room-temperature TRPL for (a) polar
80
(0001) nano-ELO GaN film on sapphire and (b) semipolar non-coalesced semi-poar GaN layer on Si. Table 3-3: PL decay times and amplitude ratios obtained from biexponential fits.
Excitation (kW/cm2)
τ1 (ns) τ2 (ns) A1/A2 τ1 (ns) τ2 (ns) A1/A2
polar c-plane nano-ELO layer (1 101) non-coalesced layer
0.02 0.16 0.54 2.02 0.33 1.03 1.38
0.06 0.19 0.58 0.65 0.38 1.42 1.12
0.13 0.15 0.64 0.22 0.38 1.68 1.10
0.19 0.16 0.71 0.15 0.38 1.80 1.11
0.25 - 0.84 0 0.39 1.90 1.17
0.32 - 1.15 0 0.40 1.98 1.19
81
Chapter 4 Quantum efficiency enhancement in GaN-based VCSELs
4.1 Hybrid VCSELs
4.1.1 Brief introduction to hybrid VCSELs
Similar with widely used GaAs-based VCSELs, GaN-based VCSELs are expected to
show various advantages over the edge-emitting lasers. Unfortunately, several obstacles
make the progress of GaN-based VCSELs relatively slow. The key problems limiting the
development of VCSELs are the difficulty in growing high quality and high reflectivity
distributed Bragg reflectors (DBRs) and to obtain high quality cavity active region which
is embedded within two highly reflectivity DBRs. High quality and high reflectivity
DBRs have a critical role in order to achieve threshold condition in VCSELs due to the
relatively short gain region. 87 In general, there are three types of III-Nitride DBRs,
including AlN/GaN, AlxGa1-xN/AlyGa1-yN, and AlxIn1-xN/GaN DBRs. The AlN/GaN
DBRs have the highest refractive index contrast (2.185 for AlN and 2.512 for GaN at
wavelength 430 nm) among all the three kinds of DBRs and provide high reflectivity
together with a relatively large stop-band width. However, the large lattice mismatch
between AlN and GaN results in accumulation of strain during growth and formation of
cracks. These cracks seriously deteriorate the reflectivity of the DBRs in addition to the
quality of the following active region. Unlike the AlN/GaN DBRs, which suffer from
cracking, the strain can be reduced in AlxGa1-xN/AlyGa1-yN DBRs with appropriate
adjustment of Al/Ga composition. However, with AlxGa1-xN/AlyGa1-yN DBRs, the
refractive index contrast decreases between the pairs of layers which results in a reduced
82
stop-band width and thus the necessity for increased number of pairs for high reflectivity.
In contrast, lattice matched Al0.83In0.17N/GaN DBRs, which could avoid subsequent layer
degradation, were reported with cavity quality factors up to 6400.88 However, relatively
lower refractive index contrast (7-8%) in Al0.83In0.17N/GaN DBRs would result in
narrower stop-band width and thereby increased number of pairs for high reflectivity.89
Besides the three kinds of III-Nitride DBRs, another possible solution for VCSELs is to
use vertical cavity with bottom dielectric DBRs instead of bottom AlN/GaN DBRs.
However, the typical employment of bottom dielectric DBRs of VCSEL requires the
removal of majority of GaN substrate with a precise control of remaining cavity length,90
which is not practical for wide application. In addition, the VCSELs performance also
depends strongly on the active region design and the promising InGaN active region
design, optimized from LEDs experiments, would be employed in VCSEL structures to
check the cavity performance. Therefore, the investigation of high quality and high
reflectivity DBRs growth and followed high quality active region growth is imperative
for high performance VCSEL devices.
4.1.2 Experimental procedures
In order to obtain crack-free AlN/GaN DBRs favorable for high reflectivity and wide
stop-band for GaN-based VCSELs, MOCVD growth for AlN/GaN DBRs on GaN/c-
sapphire template was investigated. For the purpose of releasing strain accumulation in
layers growth, 5.5 pairs AlN/GaN super-lattice layers were inserted in every 5.5 periods
of GaN/AlN DBRs deposition. The whole growth in MOCVD chamber was kept at low
pressure (76 Torr) in N2 environment. 91 The reflectivity of obtained crack-free 29 pairs
AlN/GaN DBRs and cross-sectional image of AlN/GaN DBRs from JEOL scanning
83
electron microscope (SEM) is shown in Figure 4.1. Because the DBRs is formed of AlN
and GaN layers having large refractive index difference, the reflectivity above 97% with
around 18 nm stop band with central wavelength at 435 nm was achieved with 29 pairs of
AlN/GaN DBRs.
400 450 500 5500
20
40
60
80
100
(a)
Ref
lect
ivit
y (%
)
Wavelength (nm)
Figure 4.1: (a) Reflectivity of 29 pairs crack-free AlN/GaN DBRs on GaN template, (b) cross-sectional images of 29 pairs crack-free AlN/GaN DBRs from scanning electron microscope.
The VCSEL structure, shown in Figure 4.2, contains the two active regions embedded
between the top and bottom DBRs, with each of the active regions consist of hexa 3 nm
84
In0.15Ga0.85N DHs separated by 3 nm In0.01Ga0.99N barriers. The two active regions were
separated by 132 nm (1 λ) In0.01Ga0.99N underlying layer. The designed cavity length is
2.5 λ at 410 nm for the VCSEL structure. The VCSEL structure was completed with 13
pairs SiO2/SiNx dielectric DBRs as the top DBRs, deposited at 300 °C using ultra high
vacuum chemical vapor deposition (UHVCVD). High reflectivity (> 99.5%) with ~80 nm
stop-band width was achieved with a designed stop-band central wavelength of 430 nm
for top dielectric DBRs.
GaN (~2µm)
GaN
In 0.01 Ga 0.99 N (~132 nm)
GaN
6x3nm In 0.15 Ga 0.85 N DH
6x3nm In 0.15 Ga 0.85 N DH
29 pairs AlN/GaN DBR
13.5 pairs SiO 2 /SiN x DBR
c - plane sapphire
2.5
λ
Figure 4.2: The structure schematic of the vertical cavity surface emitting laser: 2.5λ cavity with two hexa 3 nm In0.15Ga0.85N DHs separated by 132 nm (1λ) In0.01Ga0.99N underlying layer grown on bottom 29 pairs crack-free AlN/GaN DBRs on 4 µm GaN template and completed with 13 pair top SiO2/SiNx DBRs.
4.1.3 Results and discussions
Figure 4.3 shows the reflectivity spectrum for the full vertical cavity (black), reflectivity
spectrum for the bottom AlN/GaN DBRs (red) and PL spectrum (blue) emitting between
top dielectric and bottom semiconductor DBRs. The micro-PL was measured using a
85
frequency-doubled Ti:Sapphire laser (380 nm wavelength). The inset in Figure 4.3
shows the micro-PL for the cavity mode having Q-factor of 1300. However, not all the
regions on this sample showed this high Q-factor, which results in part from the non-
uniform active region quality on the wafer. During AlN/GaN bottom DBRs growth,
owing to its large lattice mismatch, a large density of dislocations was generated on the
wafer which greatly degrades the active region quality. In order to achieve higher Q-
factors, further optimization efforts focused on achieving high reflectivity bottom DBRs,
improving the active regions quality, and achieving sharper cavity interfaces were
necessary.
400 450 500 5500.0
0.2
0.4
0.6
0.8
1.0
ful cavityreflectivity
bottom DBRreflectivity
Ref
lect
ivit
y (a
rb.u
nits
)
Wavelength (nm)
PL
Figure 4.3: The reflectivity spectrum (black) for the full vertical cavity, reflectivity spectrum for the bottom AlN/GaN DBR (red) and photoluminescence spectrum (blue).
As performed earlier for the semipolar InGaN active region grown on noncoalesced GaN
layer on patterned Si, in order to gain insight into the distribution of active region
emission from the full cavity, room temperature NSOM was performed (Figures 4.4).
409 410 411
Q-factor1300
PL
Int
. (ar
b. u
nit
s)
Wavelength (nm)
86
Figures 4.4 (a) represents the AFM image of the full cavity (13 pairs top dielectric DBR
surface) for the scan area 50x50 µm2 and Figures 4.4 (b) represents the corresponding
active region emission for the scanned area. The PL emission from the microcavity is
found to be not uniform over the scanned region. The reason might be different structural
quality or differences in the cavity length in this region.
Figure 4.4: Room temperature NSOM results for the microcavity structure with bottom semiconductor and top dielectric DBRs (a) height image (b) PL intensity mapping.
87
4.2 GaN-based VCSELs with both dielectric DBRs
4.2.1 Motivation
The InGaN/GaN laser structure with bottom AlN/GaN DBRs suffers from the
degradation of active region quality. Dislocations generated during the AlN/GaN DBRs
growth would greatly degrade layer quality of the active region and result in current
leakage under electrical injection, which would severely affect laser performance. One
possible solution is to use vertical cavity with both dielectric DBRs instead of hybrid
cavity with bottom AlN/GaN DBRs. However, the typical employment of dielectric
DBRs in both sides of VCSEL cavity requires the removal of majority of GaN substrate
with a precise control of remaining cavity length. Therefore, in order to obtain high
quality of active region with high reflectivity dielectric DBRs, epitaxial lateral
overgrowth (ELO) of GaN using dielectric DBRs (SiO2/SiNx masks) as the bottom
reflector instead of the AlN/GaN bottom DBRs has been proposed. The “wing region”
(overgrown GaN layer) would have much lower density of dislocations compared with
“window region”, which would greatly improve the active region quality of the laser
cavity. Moreover, due to the larger refractive index difference between SiO2 and SiNx,
higher reflectivity (more than 99.5%) and much wider stop band (more than 80 nm) could
be achieved easily for the dielectric DBRs compared with the AlN/GaN DBRs, which
could benefits the laser cavity design. Therefore, GaN-based VCSELs with both
dielectric DBRs are promising for improvement of VCSEL performance.
88
350 400 450 500 5500.0
0.2
0.4
0.6
0.8
1.0
Ref
lect
ivit
y (a
rb.u
nits
)
Wavelength (nm) Figure 4.5: The reflectivity spectrum of bottom 13 pair SiO2/SiNx bottom DBR.
4.2.2 Experimental procedures
4.2.2.1 Growth procedures of ELO-GaN for VCSEL For the GaN-based VCSELs with both dielectric DBRs, it the critical to achieve
appropriate ELO wings for VCSELs with both dielectrics DBRs. For the purpose of
devices fabrication for electrical injection, the window width is required to be much
narrower compared with separation width, so that much longer ELO wing size could be
achieved. Nevertheless, for the ELO growth with larger fill factor (wing-to-window ratio),
the tilt of surface of the ELO-GaN with respect to underlying GaN layer becomes more
apparent, which would pose a major obstacle for the VCSEL structure. Therefore, it is
desirable to maximize the lateral versus vertical growth ratio and reduce wing tilt of the
ELO-GaN. A large lateral growth rate and a horizontal wing with greatly reduced wing
tilt were achieved by employing the NH3 flow modulation technique. In this method,
89
MOCVD chamber growth pressure was kept at 30 Torr to improve the lateral growth
ratio of GaN. The NH3 flow was interrupted for 15 s after every 20 s of regular GaN
growth, while trimethylgallium (TMGa) flow rate was kept constant. This growth was
performed using a dielectric DBRs mask with 4 μm-wide stripe-shaped windows
separated by 32 μm and aligned along the GaN 1100 direction to enhance the lateral
growth ratio. The cross-sectional images of the ELO-GaN stripes were measured with
SEM, shown in Figure 4.6.
Figure 4.6: Cross-sectional SEM images of ELO-GaN grown (a) with constant NH3 flow for 3 hours and (b) with NH3 flow modulation for 2 hours. Wing tilt angle in case of NH3 flow modulation was measured to be lower than 0.1º whereas that in the case of constant NH3 flow was 5º. From Figure 4.6, it is clear that ELO-GaN grown with NH3 flow modulation had larger
lateral growth rates and much flatter surfaces than those with constant NH3 flow. Lateral
advancement of the ELO wings with NH3 flow modulation for 2 hours was large enough
(13 μm) for the device fabrication, while the ELOs with constant NH3 flow had the same
lateral size after a longer 3 hours growth. Moreover, the modulation method resulted in a
much smaller wing tilt (< 0.1º) than the standard one (5º). This improvement is due to
enhanced migration of Ga atoms on the (0001) plane of ELO-GaN surface, since the Ga
atoms are more freely to move toward wing region with pulsed NH3 flow profile. For the
further improvement of the lateral vs. vertical growth ratio of ELO-GaN, effects of
90
growth parameters, such as temperature, NH3 flow rate, TMGa flow rate, need be
investigated in order to achieve appropriate ELO-GaN for laser diodes.
As reported in the first annual report, in order to get quality improvements of active
region, epitaxial lateral overgrowth (ELO) of GaN using dielectric DBR masks as the
bottom reflector is worthy for investigation. The “wing region” (overgrown GaN) would
have much lower density of dislocations compared with the window region. In terms of
efficiency as well as device application, it is critical to maximize the lateral versus
vertical growth ratio and achieve a flat surface and wide enough for ELO-GaN wing
region. To achieve the goal, NH3 flow modulation technique is employed to enhance the
lateral overgrowth, since it could improve the migration of Ga atoms on the (0001) plane
of ELO-GaN surface. In this method, interruption of NH3 flow was inserted for a given
time (15, 25, 30 s were test) during growth for every period (20 s) of normal GaN growth,
while TMGa (trimethylgallium) flow rate was kept constant. This growth was performed
using a growth mask with varied width such as 2, 3 and 4 μm-wide stripe-shaped
windows separated by 34, 33 and 32 μm and aligned along the GaN 1100 direction,
respectively. We tried to optimize the growth parameter such as periods of ammonia
on/off time and window size for ELO stripe to get the improvement of ELO-GaN growth
results. From the cross-section image of scanning electron microscopy (SEM) in Fig. 22,
we successfully improve the ELO-GaN lateral/vertical ratio from 1 in Figure 4.7 (a) to
around 4 Figure 4.7 (b) after 1.5 h growth.
91
Figure 4.7: Cross section pictures of ELO growth for (a) ammonia on/off time 20 s/15 s with TMGa: 20 sccm and (b) ammonia on/off time 20 s/25 s with TMGa: 12 sccm
4.2.2.2 Etching procedures for ELO-GaN for VCSEL After the optimization of the ELO-GaN growth conditions, effectively etching of ELO-
GaN layer would be employed to reduce the ELO thickness in order to achieve an
appropriate cavity length. Due to the chemical inertness and relatively high bond energies,
GaN materials are resistant to conventional wet etching, and available wet etches process
is rarely attainable with photolithography. Therefore dry etching techniques with high
plasma density and ion energies, such as reactive ion etch (RIE) or inductively coupled
plasma (ICP) etching process based on different gas mixtures and conditions, are
essentially important and extensively used for the etching of GaN films and GaN-based
device structures. Usually for normal GaN film’s etching, high etch rates and anisotropic
profiles for making mesa-patterns could be achieved if the etching conditions are
precisely controlled. Although this is convenient for mesa-pattern fabrication, the
problem of the etching of ELO-GaN has so far not been sufficiently explored in the
literatures. Variations in the quality of the window region (seed GaN) and wing region
(overgrown GaN) of ELO-GaN pattern present unique challenges for dry etching
92
processes of ELO-GaN. Therefore, it is necessary to investigate the etching conditions to
reduce ELO-GaN thickness efficiently and achieve smooth ELO-GaN surface. In this
section, we investigate the characteristics of the etching process of ELO-GaN by using a
Cl2/Ar/SiCl4 gas mixture, for potential application in fabricating ELO-GaN based
VCSEL and polariton laser structure. The use of the Cl2 for ICP etching of GaN implies
an enhancement of the chemical component to get a high etch rate. The use of Ar is for
physical etching component to balance the chemical etching component. SiCl4 serves as a
passivation layer on the etched side walls. The crystal quality of the ELO-GaN was
characterized by X-ray diffraction (XRD). The effects of etching conditions were
investigated by scanning electronic microscopy (SEM) and Atomic force microscopy
(AFM). In this work an optimization of an ICP dry etching method is made to get smooth
surface, minimum damage, a relatively high etch rate and appropriate profiles.
Regular 4 μm GaN template with a low-temperature GaN buffer layer was grown
MOCVD system on (0001) sapphire. After template growth, SiO2/SiNx DBRs (center
wavelength target at ~410 nm and stop-band ~80 nm with reflectivity >99.9%) were
deposited by PECVD on the GaN template as bottom reflector. Fabrications were applied
to open the stripe patterns along the ]0011[
direction of the underlying GaN template. The
window widths were 4 μm with a periodicity of 36 μm. ELO-GaN was grown overlying
the dielectric mask with pulse NH3 treatment and constant TMG flow. After growth, the
ELO-GaN layers had a flat plane (0001) and didn’t coalescence. The tilt of the c-axis on
the mask area was investigated by Philips X’Pert-MPDTM X-ray rocking curves (XRCs).
The XRCs of the (0002) reflection of the ELO-GaN layer was measured at ϕ = 90º, where
ϕ is the azimuth angle between the stripe direction of the mask and the scattering plane.
93
Then the ELO-GaN sample was dry etched by ICP. The etchant gases were a
Cl2/Ar/SiCl4 gas mixture. All the gases were introduced into the reactor chamber through
independent mass flow controllers (MFCs) that can control the flow rate of each gas. The
flow rates of Ar and SiCl4 were maintained at 18 and 5 sccm, respectively. The chamber
pressure during etching was 0.6 Pa. The ICP power was set at 200 W. The flow rate of
Cl2 and the bias power were modified to optimize the etching condition. The different
etching conditions were shown in Table 1. To verify the results of the dry etching, etched
surface morphology and etching anisotropy were examined by a JEOL JSM-6060 SEM.
Table 4-1: The ICP etching conditions Condition Press
(Pa)
PICP (W) PBios (W) Cl2
(sccm)
Ar (sccm) SiCl4
(sccm)
I 0.6 80 30 30 18 5
II 0.6 200 45 15 18 5
III 0.6 300 55 15 18 5
IV 0.6 300 250 15 18 5
V 0.6 80 15 30 18 5
After the ICP etching, the ELO-GaN was put into MOCVD chamber for the third time to
grow the GaN/InGaN multiple quantum wells (MQWs) and GaN cap layer on it. Veeco
DI 3100 AFM was used to investigate the surface morphology. After the growth of
MQWs and GaN cap, the top SiO2/SiNx DBR were grown by PECVD. The cross section
of the structure was observed by SEM examination.
94
Figure 4.8: The SEM cross-section image of ELO-GaN grown with pulse NH3 treatment. Figure 1 (a) shows the SEM cross-section image of ELO-GaN growth. The thickness is
about 11 μm. It is clear that with pulse NH3 treatment in ELO-GaN growth, the tilts were
greatly reduced and ELO-GaN stripe is perfect rectangular in cross-section. It gives a
very good agreement of the top DBR and the bottom DBR for VCSELs and polarition
lasers.
The ICP with Cl2/Ar/SiCl4 gas mixtures was used for the etching of ELO-GaN patterns.
The GaN etching process produces chemical byproducts such as GaCl3 which is volatile
and was pumped out. Smooth surface after dry etch of ELO-GaN is essential to
subsequent growth of good quality of device structure. This requires not only overcoming
the strong nitride bond energy but also adjusting the process conditions to deal with
inherent defects in window region of ELO-GaN. While the defects in the window region
appear to be particularly sensitive to etching conditions and respond by etching faster or
slower than the surrounding wing region of ELO-GaN. Unacceptable damaged surface or
surface morphologies with pits or peaks formation will result if the balance between the
physical and chemical aspects of the process is not maintained[15]. So in order to get good
surface, it is needed to balance the physical and chemical components during ICP etching.
95
While such balance appear some kind of sensitive. Another thing is the edge effect during
the etching of the ELO-GaN patterns. An un-optimized etching process can also result in
the peaks or facets near the edge.
Figure 4.9: (a) cross-section SEM image and (b) top-view SEM image of ELO-GaN after etching with pyramids with etching condition I. First, we tried ICP etch parameter same as what we did for ~400 nm mesa etch in LEDs
and HFETs fabrication, which is shown in condition I in Table 1. The total etching
thickness of ELO-GaN was around ~6 μm. Figure 2 are the images of etched ELO-GaN
patterns with two kind of un-optimized etching process for reference. Fig. 2 shows the
cross-section and top-view SEM image of ELO-GaN after etching with pyramids along
the stripes. Such pyramids seem large when compared with the size of ELO pattern. The
reason of the formation of the pyramids maybe that the chemical component of the
etching is much lower than the physical component. Moreover, we have to notice that the
dielectric DBRs were also etched during relatively low power ICP etch.
Then the bias power was increased to 200/45 mW and the flow rate of Cl2 was decreased
to 15 sccm to enhance the physical component of the etching to balance the chemical one.
Figure 4.10 shows the SEM image of ELO-GaN after etching with condition II in Table
1.
96
Figure 4.10: (a) cross-section SEM image and (b) top-view SEM image of ELO-GaN after etching with pyramids with etching condition II. It can be seen that there was no pyramids in the window region which means that the
enhanced physical component successfully removed pyramids which were generated
during low power etch. Both the flat window and wing regions can be obtained with this
etching condition. And the whole pattern region looks like very smooth. However, the
peaks near the edge region appeared after etching. We believe that the formation of peaks
near the edge corresponds to that the dielectric material was sputtered during the etching.
Since we already observed that dielectric DBRs were etched in low power ICP etch with
condition I, the etching rate of dielectric should be increased with higher ICP power. The
etched dielectric material was sputtered during etching, and the sputtered dielectric
material deposited again near the edge of ELO-GaN pattern. Such deposited dielectric
material near the edge will act as the masks in the subsequent etching of GaN. In this way
the peaks become larger and larger.
97
Figure 4.11: Cross-section SEM image I and (b) cross-section SEM image II of ELO-GaN after etching with pyramids with etching condition III.
When the physical component in the etching dominates more largely, the sputtered
dielectric material cannot deposit near the edge, which is shown in Figure 4.11 with etch
condition III. It is clear that the peaks at edge disappeared when we further increased
etching power. However, non-uniform etching is observed across the ELO-GaN layer,
where the center etching rate is relatively higher than the edge. This non-flat ELO-GaN
surface is an obstacle for the following cavity growth.
Figure 4.12: Cross-section SEM image and (b) top-view SEM image of ELO-GaN after etching with pyramids with etching condition IV. To overcome the non-uniform etching rate, we further increased the ICP and bias power
to 300/250 W. From Figure 4.12 (a), we can see that after long time etching, there are no
pyramids or peaks observed on sample surface. Moreover, the ELO-GaN surface is
98
relatively smooth. The DBRs without ELO-GaN covering were completed etched away,
and the bottom GaN beneath them was also removed. The edge of ELO-GaN wing is
very sharp, indicative of high physical component in ICP etching. In Figure 4.12 (b), we
can see low density of pits appeared in the window region of ELO-GaN. Unlike the wing
region of ELO-GaN, which is nearly dislocations free, the window region of ELO-GaN is
with high density of dislocations. During high power ICP etch, pits appeared due to the
high dislocations density. However, it would not affect our cavity design, since the cavity
would be deposited only in wing region of ELO-GaN.
After the ELO-GaN was etched to around 400 nm (the center of the pie-shaped sample)
and the peaks near the edge were removed, InGaN active region and p-GaN layer were
grown on the etched ELO-GaN. The thickness of the structure between top and bottom
DBR is designed to be around 1.5~2 μm. This structure is hopeful for the application in
the VCSEL structure.
4.2.2.3 Full cavity VCSELs After reducing the ELO-GaN thickness down to 400 nm using ICP etching, the sample
was loaded into the MOCVD chamber and subjected to an in situ H2 treatment at high
temperature (950 ºC) to remove the residual etching damage on ELO-GaN surface. Since
multi-DHs have exhibited a superior approach for quantum efficiency enhancement
ompared to MQW as LED active regions, multi-DH active regions were employed in
VCSEL cavity, shown in Figure 4.13 (a). The cavity active region features 2 periods of
hexa 3 nm-thick In0.13GaN DH (separated by 3 nm-thick In0.01GaN barriers) separated by
160 nm (λ) In0.01GaN underlying layer. The designed emission wavelength was 410 nm.
99
Here two periods of hexa 3 nm In0.15GaN DHs were employed in order to feed to cavity
more photons to obtain better cavity mode. In addition to vertical cavity sample a test
sample was simultaneously grown on c-plane GaN on sapphire substrate in order to
compare the optical quality of the vertical cavity active region. Similar pulsed-NH3
method as ELO-GaN growth was employed for p-GaN growth to ensure p-GaN would
spread SiO2 layer, where the p-contact would be formed. The full vertical cavity was
finished with 13.5 pair SiO2/SiNx top DBRs deposited using PECVD system with target
stop band wavelength around 400 nm.
Figure 4.13: (a) the VCSEL full vertical cavity structure on c-sapphire. (b) Electric field inside the cavity with respect to distance where an InGaN active region is placed at the antinode of the electric field inside the cavity. It should be noted that a 300 nm-thick SiO2 layer was deposited using PECVD system
followed with a photolithography procedure after n-GaN growth. 300 nm SiO2 layer was
deposited on ELO-GaN layer after n-GaN growth. Then photolithography pattern was put
100
on the oxide layer, followed by buffer oxide etcher (BOE) wet etching. After etching, 8
um wide SiO2 stripes separated by 10 um were achieved with the same orientation as
ELO stripes. The window regions of ELO-GaN and separation region between ELO-GaN
stripes were covered with SiO2 and will not participate in current injections. Therefore,
the SiO2 layer ensures the electrical injection goes through InGaN active region grown on
dislocation free ELO-GaN wing region only. Figure 4.13 (b) demonstrates the multi-DH
active regions are designed to place on electric field antinodes in the cavity in order to
maximize the cavity emission. However, due to the thickness variation of the ELO-GaN
layers beneath the active regions, the placement of the active regions within the cavity
varies across the sample. Further optimization to form homogeneous ELO GaN layers is
necessary for better optimized vertical cavity structures.
4.2.3 Optical characterization for VCSEL
The VCSEL full cavity structure was characterized using reflectivity (Xe lamp source)
for the reflectivity spectrum in the full cavity structure, shown in Figure 4.14. It is clear
that 99% reflectivity is achieved for the top SiO2/SiNx DBR with a stop-band width of ~
80 nm, which is much larger compared with that of semiconductor AlN/GaN DBRs (~18
nm). The full-cavity reflectivity spectrum showed clear dips corresponding to the cavity
modes at ~ 400 nm and ~ 412 nm.
101
340 360 380 400 420 440 460 4800
20
40
60
80
100
390 400 410 420
96
97
98
99
100
Reflect
ivity
(%)
Wavelength (nm)
Cavity modes
Re
flect
ivity
(%
)
Wavelength (nm)
Figure 4.14: Reflectivity spectrum of the full cavity structure. Cavity modes are observed around 400 and 412 nm can be clearly seen in inset.
Micro-PL measurement was employed to get the Q factor of the VCSEL sample as well
as the simultaneously grown reference sample. The samples were excited normal to the
surface from the top DBRs side using a HeCd laser with wavelength 325 nm. The 325 nm
wavelength excitation ensured to excite the active region outside of the top DBRs stop
band from 350 nm to 430 nm (see Figure 4.14). The emission was also collected normal
to the surface from the top DBR side. The excited area was in the order of ~ 2 m in
diameter. Figure 4.15 shows the PL spectra for the full cavity structure, half cavity
structure and reference sample. The half cavity Q-factor is 60, 3 times higher than
reference sample Q-factor 20, clearly showing the confinement of PL spectrum from
bottom DBRs. The micro-PL for the cavity mode has an average Q-factor of 500 across
102
the wafer, which was substantially higher compared with half cavity Q-factor after the
top DBRs deposition. The result proves the obtained high quality full cavity structure,
which is very promising for the electrical operation. However, we have to note that the
vertical cavity thickness varies across the wafer. Owing to non-uniform vertical thickness
of ELO-GaN after MOCVD growth, during ICP-etching on ELO-GaN, some portion of
ELO-GaN stripes remained thickness of a few micro while some portion of ELO-GaN
stripe was nearly etched away.
390 400 410 420 430 440 450 460
Active region (reference)Q-factor: 20
Full cavityQ-factor: 500
PL Inte
nsi
ty (
arb
. units
)
Wavelength (nm)
Half cavityQ-factor: 60
Figure 4.15: PL spectra for the full cavity structure (red), half cavity structure (black) and reference sample (blue).
4.2.4 Fabrication and electrical characterization for VCSEL
Fabrication was processed on the full cavity for electrical characterization. One challenge
for the GaN-based VCSEL fabrication with both dielectric DBRs is current confinement.
Since the ELO-GaN window region has a high density of dislocations, which could be
103
current leak path. It is necessary to do a selective oxidation to prevent current spreading
to window region. 300 nm SiO2 layer was deposited after n-GaN growth followed with
photolithography pattern. 8 um wide SiO2 stripes separated by 10 um were achieved with
the same orientation as ELO stripes to cover SiO2 layer on the window regions of ELO-
GaN and separations between ELO-GaN stripes. Therefore, only InGaN active region
grown on high quality ELO-GaN wing region will participate in current injections. To
form the n-contact, the top DBRs and the whole active regions were etched away by ICP
dry etching. Ti/Al/Ni/Au (30/100/40/50 nm) were served as n-contact, followed with
RTA annealing at 860oC for 1min. Pulse-NH3 method was also employed for p-GaN
growth to ensure p-GaN would spread SiO2 layer, where the p-contact would be formed.
Ni/Au (5/5 nm) was served as p-contact with Ni/Au (40/50 nm) served as p-pad for
probes. The full schematic of fabricated devices is shown in Figure 4.16.
Figure 4.16: Cross-sectional schematics of fabricated VCSEL device with electrical contacts.
The mask designs to produce the vertical cavity devices with different sizes using
photolithography and e-beam deposition techniques are shown in Figure 4.17. Figure
4.17 (a) illustrates the mesa patterns on ELO-GaN stripes, while Figure 4.17 (b) shows
104
the final device shapes with contact layers. The devices are prepared to constitute on
changing mask patterns in the dimensions of 540x540, 200x200, 20x40, 40x20, 20x20,
20x10 μm2. The p-contacts are shown in black stripes, while p-pad (red net) was
deposited on top of dielectric DBRs. N-contacts (Ti/Al/Ni/Au) deposited on regions
between mesa patterns.
Figure 4.17: VCSEL devices on the mesa, (b) the final device shapes with contact layers.
a) b)
105
Chapter 5 Conclusions and future research
5.1 Conclusions
InGaN based LEDs have been widely used in displays of TV, computers, and cell
phones. And more applications have been focused on the general lighting market due to
their high efficiency and long lifetime. However, high power LEDs suffer efficiency
droop at high injection levels. The theoretical and experimental results along with work
done by other researchers have ruled out the possibility of Auger recombination in wide
band gap GaN based LEDs. We have proposed that the poor hole transport in GaN is a
obvious contributing factor for such efficiency droop. Delta-doping with Mg of barriers
significantly improves the quantum efficiency of LEDs with active regions composed of
hexa 3-nm DHs. With careful design of the active regions, the significant improvement in
hole injection was achieved without degradation of the material quality due to the
presence of Mg. With Mg delta-doping of the first barrier on the LED n-side, the relative
peak EQE improves by 20% in compared with the reference structure. Doping of the first
and second barriers on the n-side does not result in further improvement of relative EQE
compared to the case of one doped barrier. This possibly indicates that further
improvement in hole injection is canceled by the reduced active region quality due to
higher Mg-doping. The both LED structures with Mg delta-doped barriers effectively
suppress the efficiency droop at high injection level compared to the counterpart with un-
doped barriers.
Armed higher quantum efficiency in our LED structures, we investigated various multi-
DH active region designs to maximize the light output. By using multiple thin DH
106
structures, the relative EL efficiency was enhanced greatly by increasing number of DH
up to hexa 3 nm DH. It is clear that with more DHs in the active region (less than 6), the
LED can handle more injection power and thus favors high power applications. Our
optimization of SEI structures increasing SEI thickness from 5 nm to 30 nm per step
further improved relative EL efficiency for single 3 nm DH and move the peak efficiency
of 3 nm DH to the comparable level with quad 3 nm DH albeit single 3 nm DH shows ~2
time lower EL efficiency than quad 3 nm DH at high injection levels. It is shown that a
good design of SEI may play an important role for the efficiency improvement of LEDs,
especially for LEDs with thin active layers.
Novel pulsed-NH3 growth techniques have been used for GaN-based VCSEL structure
with growing active layers on nearly defect free epitaxial lateral overgrown (ELO) GaN
layers embedded in bottom and top dielectric DBRs. ELO-GaN growth provides a
superior solution to VCSELs. While VCSEL on bottom AlN/GaN DBRs suffers narrow
stop-band and high density of defects, resulting quality degradation of active layers,
ELO-GaN growth technique could provide wider stop-band dielectric DBRs and high
quality InGaN active regions. Quality factors up to 1300 was obtained for the VCSEL
structure with dual hexa 3 nm DH active region on semiconductor bottom DBRs. In order
to achieve the appropriate ELO-GaN for vertical cavity structure, MOCVD growth
parameters was studied for higher lateral to vertical (L/V) growth ratio for ELO-GaN
wings. Furthermore, the etching parameters to thin down ELO-GaN using inductively
coupled plasma (ICP) technique was investigated to obtain desired vertical cavity lengths
and reduce defects density after etching. Final cavity structure was obtained after the
lithography techniques used following with e-beam deposition for the electrical contacts.
107
5.2 Future research
From aforementioned discussions supported by massive theoretical and experimental
results, we can conclude that in order to further mitigate efficiency droop and improve
the GaN-based LEDs, hole concentrations must be further increased. However, standard
GaN growth conditions by MOCVD lead to hole concentration typically on the order of
4-7×1017 cm-3 since the Mg activation energy in GaN is high and the formation of the
Mg-H complexes inhibit the ionization of Mg acceptors. Using higher growth pressure
and lower temperature can enable higher hole concentrations.92 However, low growth
temperature leads to GaN material quality degradation9392. In addition, this approach is
limited by self-compensation attributed to nitrogen vacancy complexes in p-GaN. Due to
the motion of Fermi level, nitrogen vacancies are expected to have a major impact on p-
GaN. However, with higher NH3 partial pressures during growth and careful post
annealing, the nitrogen vacancy density can be minimized. For p-GaN growth with Ga
polarity, the Mg incorporation can induce the stacking faults from GaNGaN to
GaNMgNGa, inverting the GaN polarity from Ga-face to N-face. 94 As the growth
proceeds, additional Mg atoms migrate towards these stacking faults leading them to
develop along several inclined planes, and eventually form pyramidal-shape defects,
called pyramidal inversion domains. Delta-doping of Mg into GaN was found to be
virtually free of such inversion domains extended defects due to the hindering of the
vertical diffusion of Mg inhibited by GaN interlayers, which in turn improved the surface
morphology93. It was also reported that Mg delta-doping improves not only the p-type
GaN conduction, but also significantly suppresses the dislocation densities, which is
108
beneficial to the leakage current and lifetime of LEDs.95 It has also been demonstrated
that incorporating Mg-doped AlGaN/GaN superlattice structure into devices could
enhance the hole conduction in the lateral direction. However, the enhancement of hole
conduction in the vertical direction by employing a superlattice structure is limited
because a superlattice structure simultaneously introduces potential barriers for hole
conduction in the vertical direction.
Figure 5.1: delta- doping profile implementation: Stage I: undoped GaN growth (u-GaN); II: Nitridation; III: Mg incorporation (constant Mg molar flow for all the samples); Using the Mg δ- doping technique, hole concentration as high as 1018 cm-3 has been
achieved by MOCVD growth method96. We expect that, by tuning the growth parameters
(including growth temperature, Mg flow rate, interruption time and undoped GaN spacer
layer thickness) during growth of Mg δ- doped GaN, the solubility limit of Mg into GaN
can be affected in a controlled way, allowing to enhance the incorporation of Mg ions, to
hinder the self-compensation mechanisms and therefore improve the concentration of
active carriers in the layers. For p-type doping studies, a p-n--n+ layer structure was
grown. The growth conditions for top Mg δ- doping GaN have been varied by changing
the undoped GaN spacer thickness from 5 nm, 7.5 nm, 10 nm, to 12.5 nm. The bottom n
109
type layers were Si-doped with carrier concentration around 3×1018 cm-3. This was
capped with lightly Si doped GaN (5×1017 cm-3) layer of 200 nm thickness, and finally
500 nm Mg δ- doping GaN. This structure is useful for making electrical measurements
on Mg doped layer because the resulting p-n junction has a high reverse breakdown
voltage; as a consequence a high bias can be applied between two p-contacts and only
hole current will flow between them, enabling accurate measurements. In addition, by
forming p-n junction devices, we can test the I-V characteristics for the p-n junctions
simultaneously. The results are shown in Table 5-1 using TLM measurement [Figure 5.2
(a)] assuming hole mobility 5 cm2/Vs.
Table 5-1 TLM pattern measurement results of delta doped p-GaN Specific contact
resistance
(Ohm cm2)
Sheet
resistance
(Ohm)
Hole concentration
(cm-3)
Mg delta doping
(5nm)
3.5*10-1
1.2*106
0.2*1017
Mg delta doping
(7.5nm)
1.6*10-2
1.6*105
1.5*1017
Mg delta doping
(10nm)
6.6*10-3
6.3*104
4.0*1017
Mg delta doping
(10nm)
6.2*10-3
6.5*104
3.8*1017
Mg 200sccm uniform 4.8*10-3 3.3*104 7.6*1017
110
doping
4 5 6 7 8 9 10 11 12 13 14
0
1
2
3
4
5
6
(b)
Ca
rrie
r co
nce
ntr
atio
n (
10
17
cm-3
)
Thickness of u-GaN spacer (nm)
Figure 5.2: (a) Cross-sectional schematic of TLM pattern; (b) Measured hole concentration as a function of u-GaN spacer thickness
111
As presented in Error! Reference source not found. (b), with increasing u-GaN spacer
thickness up to 10 nm, the hole concentrations continue improving. The highest hole
concentration obtained so far is 4×1017 cm-3 with u-GaN thickness. Further increasing u-
GaN thickness results in reduced hole concentration probably due to decreased Mg
doping concentration. Although the obtained hole concentration is still less than our
standard continuous Mg doping p-type GaN with hole concentration 7×1017 cm-3, the Mg
delta-doping GaN optimization is still on the way. Further optimizations are required in
order to achieve higher hole concentration including (1) optimizing Mg source flow; (2)
optimizing u-GaN spacer growth V/III ratio.
112
References 1 H. Morkoç, Handbook of Nitride Semiconductors and Devices, Volume 3, Chapter 1, Wiley-
VCH, (2008)
2 http://www.nobelprize.org/nobel_prizes/physics/laureates/2014/popular-physicsprize2014.pdf
3 S. Nakamura and M.R. Krames, Proceedings of the IEEE , 101, 2211 (2013)
4 M. J. Cich, R. I. Aldaz, A. Chakraborty, A. David, M. J. Grundmann, A. Tyagi, M. Zhang, F. M.
Steranka, and M. R. Krames, Appl. Phys. Lett. 101, 223509 (2012)
5 T. Delaney, P. Rinke, and C.G. Van de Walle, Appl. Phys. Lett. 94, 191109 (2009).
6 M. H. Kim, M. F. Schubert, Q. Dai, J. K. Kim, E. F. Schubert, J. Piprek, and Y. Park, Appl.
Phys. Lett. 91, 183507 (2007)
7 Y.C. Shen, G.O. Mueller, S. Watanabe, N.F. Gardner, A. Munkholm, M.R. and Krames, Appl.
Phys. Lett. 91, 141101 (2007)
8 S. H. Park, Y. T. Moon, D. S. Han, J. S. Park, M. S. Oh, and D. Ahn, Semicond. Sci. Technol.
27, 115003 (2012)
9 B. Monemar and B.E. Sernelins, Appl. Phys. Lett. 91, 181103 (2007)
10 X. Li, S. Okur, F. Zhang, V. Avrutin, Ü. Özgür, and H. Morkoç, and K. Jarasiunas, App. Phys.
Lett. 101, 041115 (2012)
11 A. Laurain, M. Myara, G. Beaudoin, I. Sagnes, and A. Garnache, Optics Express 18 (14), 14627
(2010)
12 S. C. Wang, T. C. Lu, H. C. Kuo, and J. R. Chen, GaN based VCSEL, Chapter 13, Springer
Berlin Heidelberg (2013)
13 T. C. Lu, J. R. Chen, S. W. Chen, H. C. Kuo, C. C. Kuo, C. C. Lee, and S. C. Wang, IEEE J.
Selected Topic in Quantum Electronics, 15, 850 (2009)
14 http://spie.org/x87351.xml
15 A. Michiue, et al., Proc. SPIE. 56, 7216 (2009).
113
16 US Department of Energy, Solid-State Lighting Research and Development: Multi-Year
Program Plan (March 2010)
17 www.laserfocusworld.com/articles/print/volume-46/issue-11/features/photonic-frontiers-white-
light-leds-white-light-leds-promise-a-bright-future-for-solid-state-illumination.html
18 B. Monemar and B. E. Sernelius, Appl. Phys. Lett. 91, 181103 (2007).
19 I. V. Rozhansky and D. A. Zakheim, Semiconductors 40, 839 (2006).
20 I. V. Rozhansky and D. A. Zakheim, Phys. Status Solidi A 204, 227(2007).
21 I. A. Pope, P. M. Smowton, P. Blood, J. D. Thomson, M. J. Kappers, and C. J. Humphreys,
Appl. Phys. Lett. 82, 2755 (2003).
22 Min-Ho Kim, Martin F. Schubert, Qi Dai, Jong Kyu Kim, and E. Fred Schubert, Joachim
Piprek and Yongjo Park, Appl. Phys. Lett. 91, 183507 (2007).
23 M. F. Schubert, J. Xu, J. K. Kim, E. F. Schubert, M. H. Kim, S. Yoon, S. M. Lee, C. Sone, T.
Sakong, and Y. Park, Appl. Phys. Lett. 93, 041102 (2008).
24 Y. C. Shen, G. O. Mueller, S. Watanabe, N. F. Gardner, A. Munkholm, and M. R. Krames,
Appl. Phys. Lett. 91,141101 (2007).
25 A. A. Efremov, N. I. Bochkareva, R. I. Gorbunov, D. A. Larinovich, Yu. T. Rebane, D. V.
Tarkhin, and Yu. G. Shreter, Semiconductors. 40,605 (2006).
26 J. Iveland, L. Martinelli, J. Peretti, J. S. Speck, and C. Weisbuch, Phys. Rev. Lett. 110, 177406
(2013)
27 J. Hader, J. V. Moloney, B. Pasenow, S. W. Koch, M. Sabathil, N. Linder, and S. Lutgen, Appl.
Phys. Lett. 92, 261103 (2008)
28 N. F. Gardner, G.O. Müller, Y. C. Shen, G. Chen, S. Watanabe, W. Götz, and M. R. Krames,
Appl. Phys. Lett. 91, 243506 (2007).
29 A. R. Beattie and P. T. Landsberg, Proc. R. Soc. Lond. A. 249,1256 (1958).
30 J. Xie, X. Ni, Q. Fan, R. Shimada, Ü. Özgür, and H. Morkoç, Appl. Phys. Lett. 93, 121107
114
(2008)
31 X Ni, X Li, J Lee, S Liu, V Avrutin, U Ozgur, H Morkoç, and A Matulionis, J. Appl. Phys. 108,
033112 (2010)
32 X. Ni, X. Li, J. Lee, S. Liu, V. Avrutin, A. Matulionis, Ü. Özgür, and H. Morkoç, Superlattices
and Microstructures 48, 133 (2010)
33 K. T. Tsen, R. P. Joshi, D. K. Ferry, A. Botchkarev, B. Sverdlov, A. Salvador, and H. Morkoç,
Appl. Phys. Lett. 68, 2990 (1996).
34 I. V. Rozhansky and D.A. Zakheim, Phys. Stat. Sol. A 204, 227 (2009).
35 X. Ni, X. Li, J. Lee, S. Liu, V. Avrutin, Ü. Özgür, H. Morkoç, A. Matulionis, T. Paskova, G.
Mulholland, and K. R. Evans, Appl. Phys. Lett. 97, 031110 (2010).
36 Ü. Özgür, H. Liu, X. Li, X. Ni, and H. Morkoç, Proc. IEEE 98, 1180 (2010).
37 J. F. Muth, J. H. Lee, I. K. Shmagin, R. M. Kolbas, H. C. Casey, B. P. Keller, U. K. Mishra, S.
P. DenBaars, Appl. Phys. Lett. 71, 2572 (1997).
38 J. Y. Duboz, F. Binet, D. Dolfi, N. Laurent, F Scholz, J. Off, A. Sohmer, O. Briot, B. Gil, Mat.
Sci. Eng. B 50, 289 (1997).
39 Q. Dai, M. F. Schubert, M. H. Kim, J. K. Kim, E. F. Schubert, D. D. Koleske, M. H. Crawford,
S. R. Lee, A. J. Fisher, G. Thaler, and M. A. Banas, Appl. Phys. Lett. 94, 111109 (2009).
40 S. Jursenas S. Miasojedovas, G. Kurilcik, and A. Zukauskas, Appl. Phys. Lett. 83 (2003) 66.
41 X. Li, H.Y. Liu, X. Ni, Ü. Özgür, and H. Morkoç, Superlattices Microstructure, 47 (2010) 118.
42 M. Binder, A. Nirschl, R. Zeisel, T. Hager, H.-J. Lugauer, M. Sabathil, D. Bougeard, J. Wagner,
and B. Galler, Appl. Phys. Lett, 103, 071108 (2013)
43 J. Hader, J.V. Moloneya, and S. W. Koch, Proc. of SPIE OptoWest Symposium, Feb 3-6, Vol.
8986 San Francisco CA
44 X. Ni, Q. Fan, R. Shimada, Ü. Özgür, and H. Morkoç, Appl. Phys. Lett. 93, 171113 (2008). 45 B. Monemar and B.E. Sernelius, Appl. Phys. Lett. 91, 181103 (2007).
46 B.J. Ahn, T.S. Kim, Y. Dong, M.T. Hong, J.H. Song, Jae-Ho Song, H.K. Yuh, S.C. Choi, D.K.
115
Bae, and Y. Moon, Appl. Phys. Lett. 100, 031905 (2012).
47 A. David, M.J. Grundmann, J.F. Kaeding, N.F. Gardner, T.G. Mihopoulos, and M.R. Krames,
Appl. Phys. Lett. 92, 053502 (2008).
48 J.P. Liu, J.H. Ryou, R.D. Dupuis, J. Han, G.D. Shen, and H.B. Wang, Appl. Phys. Lett. 93,
021102 (2008).
49 N. F. Gardner, G.O. Müller, Y. C. Shen, G. Chen, S. Watanabe, W. Götz, and M. R. Krames,
Appl. Phys. Lett. 91, 243506 (2007).
50 M. Maier, T. Passow, M. Kunzer, W. Pletschen, K. Köhler, and J. Wagner, Phys. Status Solidi C
7, 2148 (2010).
51 X .Li, S. Okur, F. Zhang, V. Avrutin, S.J. Liu, Ü. Özgür, H. Morkoç, S.M. Hong, S.H. Yen, T.S.
Hsu, and A. Matulionis, J. Appl. Phys. 111, 063112 (2012).
52 J. Xie, Ü. Özgür, Y. Fu, X. Ni, H. Morkoç, C.K. Inoki, and T.S. Kuan, J.V. Foreman and H.O.
Everitt, Appl. Phys. Lett. 90, 041107 (2007).
53 S. Watanabe, N. Yamada, M. Nagashima, Y. Ueki, C. Sasaki, Y. Yamada, T. Taguchi, K.
Tadatomo, H. Okagawa, and H. Kudo, Appl. Phys. Lett. 83, 4906 (2003).
54 Y.J. Lee, C.H. Chiu, C.C. Ke, P.C. Lin, T.C. Lu, H.C. Kuo, and S.C. Wang, IEEE J. Quant.
Electron. 15, 1137 (2009).
55 J.Hader, J.V. Maloney, and S.W. Koch, Appl. Phys. Lett. 96, 221106 (2010).
56 T. Malinauskas, A. Kadys, T. Grinys, S. Nargelas, R. Aleksiejunas, S. Miasojedovas, J.
Mickevicius, R. Tomasiunas, K. Jarasiunas, M. Vengris, S. Okur, X. Li, F. Zhang, V. Avrutin,
Ü. Özgür and H.Morkoç, Proc. SPIE 8262, 82621S (2012).
57 X.A. Cao, E.B. Stokes, P.M. Sandvik, S.F. LeBoeuf, J. Kretchmer, and D. Walker, IEEE
Electron Device Lett. 23, 535 (2002).
58 X. Ni, X. Li, J. Lee, S. Liu, V. Avrutin, Ü. Özgür, H.Morkoç, and A. Matulionis, J. Appl. Phys.
108, 033112 (2010).
116
59 X. Li, S. Okur, F. Zhang, V. Avrutin, Ü. Özgür, and H. Morkoç, and K. Jarasiunas, App. Phys.
Lett. 101, 041115 (2012)
60 A. Armstrong, T. A. Henry, D. D. Koleske, M. H. Crawford, K. R. Westlake, and S. R. Lee,
Appl. Phys. Lett. 101, 162102 (2012)
61 J. Liberis, I. Matulionienė , A. Matulionis, M. Ramonas, and L. F. Eastman, in Advanced
Semiconductor Materials and Devices Research: III-Nitrides and SiC, edited by H.-Y. Cha
(Transworld Research Network, Kerala, India, 2009).
62 Q. Dai, Q.F. Shan, J.H. Cho, E.F. Schubert, M.H. Crawford, D.D. Koleske, M.H. Kim, and Y.
Park, App. Phys. Lett. 98, 033506 (2011).
63 K. J. Vampola, M. Iza, S. Keller, S. P. DenBaars, and S. Nakamura, Appl. Phys. Lett. 94,
061116 (2009)
64 H.Y. Ryu, and J. M. Lee, Appl. Phys. Lett. 102, 181115 (2013)
65 Y.K. Kuo, M. C. Tsai, S. H. Yen, T. C. Hsu, and Y. J. Shen, IEEE J. Quantum Electron. 46, 1214
(2010)
66 Y. Y. Zhang, G. H. Fan, and T. Zhang, IEEE J. Quantum Electron. 48, 169 (2012)
67 S. H. Han, C. Y. Cho, S. J. Lee, T. Y. Park, T. H. Kim, S. H. Park, S. W. Kang, J. W. Kim, Y. C.
Kim, and S. J. Park, Appl. Phys. Lett. 96, 051113 (2013)
68 I. V. Rozhansky and D. A. Zakheim, Semiconductors 40, 839 (2006)
69 V. Avrutin, S. Hafiz, F. Zhang, Ü. Özgür, H. Morkoç, and A. Matulionis, J. Vac. Sci. Technol. A
31, 050809 (2013)
70 F. Zhang, X. Li, S. Hafiz, S. Okur, V. Avrutin, Ü. Özgür, H. Morkoç and A. Matulionis, Appl.
Phys. Lett. 103, 051122 (2013)
71 Patrik Ščajev, Kęstutis Jarašiūnas, Serdal Okur, Ümit Özgür, and Hadis Morkoç, “ Carrier
Dynamics in Bulk GaN”, Journal of Applied Physics, v 111, n 2, p 023702 (8 pp.), 15 Jan. 2012
72 J.S. Im, H. Kollmer, J. Off, A. Sohmer, F. Sholz, A. Hangleiter, Mater. Res. Soc. Symp. Proc.
117
482, 513 (1997).
73 P. Perlin, C. Kisielowski, V. Iota, B. A. Weinsten, L. Mattos, N.A. Shapiro, J. Kruger, E.R.
Weber, J. Yang, Appl. Phys. Lett. 73, 2778 (1998).
74 O. Ambacher, J. Majewski, C. Miskys, A. Link, M. Hermann, M. Eickhoff, M. Stutzmann, F.
Bernardini, V. Fiorentini, V. Tilak, B. Schaff and L.F. Eastman, J. Phys. : Condens. Matter 14,
3399 (2002).
75 P. Blood, IEEE J. Quantum Elec. 36, 354 (2000).
76 V.I. Litvinov, J. Appl. Phys. 88, 5814 (2000).
77 R. J. Radtke, U. Waghmare, H. Ehrenreich, and C. H. Grein, Appl. Phys. Lett. 73, 2087 (1998).
78 A. Dmitriev and A. Oruzheinikov, J. Appl. Phys. 86, 3421 (1999).
79 Q. Dai, M. F. Schubert, M. H. Kim, J. K. Kim, E. F. Schubert, D. D. Koleske, M. H. Crawford,
S. R. Lee, A. J. Fischer, G. Thaler, and M. A. Banas, Appl. Phys. Lett. 94, 111109 (2009).
80 L. Wang, C. Liu, J.N. Lu, L. Liu, N.Y. Liu, Y.J. Chen, Y.F. Zhang, E.D. Gu, and X.D. Hu,
Optics Express 19, 14182 (2011).
81 F.D. Sala, A.D. Carlo, P. Lugli, F. Bernardini, V. Fiorentini, R. Scholz, and J.M. Jancu, Appl.
Phys. Lett.74, 2002 (1999).
82 E. Kioupakis, Q. Yan, C. G. Van de Walle, Appl. Phys. Lett. 101, 231107 (2012)
83 N. Izyumskaya, S. J. Liu, S. Okur, M. Wu, V. Avrutin, U. Ozgur, S.Metzner, F. Bertram, F.
Bertram, L. Zhou, D. J. Smith, and H. Morkoc¸, Proc. SPIE 7939, 79391W (2011)
84 N. Izyumskaya, S. J. Liu, V. Avrutin, S. Okur, U. Ozg€ur, S. Metzner, C. Karbaum, F. Bertram,
J. Christen, D. J. Smith, and H. Morkoc¸, Proc. SPIE 8262, 826224 (2012)
85 N. Izyumskaya, F. Zhang,S. Okur, T. Selden, V. Avrutin, U. Ozgur, S. Metzner, C. Karbaum, F.
Bertram, J. Christen, and H. Morkoc, J. Appl. Phys. 114, 113502 (2013)
86 S. Okur, S. Metzner, N. Izyumskaya, F. Zhang, V. Avrutin, C. Karbaum, F. Bertram, J. Christen,
118
H. Morkoç, and Ü. Özgür, Appl. Phys. Lett. 103, 211908 (2013)
87 W. W. Chow, K. D. Choquette, M. H. Crawford, K. L. Lear, and G. R. Hadley, IEEE J.
Quantum Electron. 33, 1810 (1997)
88 R. Butté, G. Christmann, E. Feltin, A. Castiglia, J. Levrata, G. Cosendey, A. Altoukhov, J. F.
Carlin, and N. Grandjean, Proc. of SPIE, 7216, 721619, (2009)
89 R. Butte, J. Carlin, E. Feltin, M. Gonshorek, S. Nicolay, G. Christmann, D. Simeonov, A.
Castiglia, J. Dorsaz, H. Buehlmann, S. Chirstopoulos, G. Hongersthal, A. Grundy, M. Mosca, C.
Pinquier, M. Py, F. Demangeot, J. Frandon, P. Lagoudakis, J. Baumberg, and N. Grandjean, J.
Phys. D: Appl. Phys. 40, 6328 (2007)
90 C. Holders, J. Speck, S. DenBaars, S. Nakamura, and D. Feezell, Appl. Phys. Express. 5,
092104 (2012)
91 G. S. Huang, T. C. Lu, H. H. Yao, H. C. Kuo, S. C. Wang, C. Lin, and L. Chang, Appl. Phys.
Lett. 88, 061904 (2006)
92 P. Kozodoy, S. Keller, S.P. DenBaars, and U.K. Mishra, J. Cryst. Growth, 195, 265 (1998).
93 T. Li, C. Simbrunner, M. Wegscheider, A. Navarro-Quezada, M. Quast, K. Schmidegg, and A.
Bonanni, J. Cryst. Growth, 310, 13 (2008).
94 S. Hautakangas, K. Saarinen, L. Liszzkay, J.A. Freitas, Jr., and R.L. Henry, Phys. Rev. B
72,165303 (2005).
95 H. Kim, J. Li, S.X. Jin, J.Y. Lin, H.X. Jiang, Appl. Phys. Lett. 83, 566 (2003).
96 C. Bayram, J.L. Pau, R. McClintock, and M. Razeghi, 104, 083512 (2008).
119
Curriculum Vitae
Fan Zhang Date of Birth: Oct. 16, 1983 Citizenship: P.R. China Email: zhangf4@vcu.com EDUCATION
Virginia Commonwealth University, Richmond, Virginia
Electrical Engineering, Ph.D., December 2014
Chinese Academy of Science, Beijing, China
Material Science and Engineering, M.S., July 2008
Peking University, Beijing, China
Microelectronics, B.S., July 2005
TECHNICAL SKILLS
Semiconductor device fabrication and processing: Photolithography/Wet Etching
/Metallization /E-beam Deposition/Sputtering/Plasma Etching /Wire Bonding
MOCVD epitaxial thin film growth
PECVD deposition of SiO2/SiNx
Device testing: C-V, I-V, AC/DC/RF, LCR meter, TLM, EL, integrating sphere
Characterizations: XRD, SEM, AFM, PL, Profilometry, Hall measurements
Computers: Matlab, JMP, Silvaco TCAD, Mathematica, Origin, MSOffice
PROFESSIONAL EXPERIENCE
Microelectronics Materials and Device Laboratory--Virginia Commonwealth University
Research Assistant, Aug. 2009 to Present
Designed InGaN active regions (single and multiple quantum wells) to improve quantum
efficiency of LEDs
Optimized electron cooler layers for efficiency droop mitigation to enhance LEDs light-
output
120
Designed novel delta p-doped quantum barrier in InGaN active regions and optimized p-
GaN structure to improve hole injection of LEDs
Simulated LED devices using Silvaco TCAD and incorporated into structure designs
Developed novel growth technique for ELO-GaN structures on SiO2/SiNx Distributed
Bragg Reflectors (DBR) on sapphire
Achieved crack- free semiconductor DBRs with reflectivity as high as 98% for vertical
cavity laser (VCSEL)applications
Developed InGaN/GaN phonon cavity on AlN/GaN superlattices
Investigated dislocations, point defects, basal and prismatic stacking faults of III-Nitride
thin film layer
Developed dry etching process to achieve smooth surface with minimized surface
damages and defects density
Demonstrated technology to obtain high quality semi-polar and non-polar GaN layers on
Si substrates
Improved HEMTs channel mobility and reduced sheet resistance
Developed novel couple/dual-channel HEMTs to reduce hot phonon effect and improve
devices reliability
Simulated HEMT structures in Silvaco TCAD and incorporated into growth designs
Developed dielectric thin films for passivation to improve HEMTs performance
Optimized plasma etching conditions and eliminated HEMT device leakage current
Reduced specific ohmic contact resistance by one order of magnitude for HEMTs
Investigated AlGaN HEMT on Si substrates
Institute of Semiconductor of Chinese Academy of Science – Beijing, China
Research Assistant, 08/05-07/08
Achieved high quality ZnO materials with chemical vapor transport growth
KEY PUBLICATIONS
F. Zhang, H. Morkoc, et al., “The effect of stair case electron injector design on electron
overflow in InGaN light emitting diodes”, Appl. Phys. Lett., 103, 051122 (2013)
F. Zhang, H. Morkoc, et al. “The impact of active layer design on quantum efficiency of
InGaN light emitting diodes”, SPIE OPTO, 82621G (2012)
121
F. Zhang, H. Morkoc, et al. “GaN-based vertical cavity lasers with
semiconductor/dielectric and all dielectric reflectors”, Proc. of SPIE, 8625, 86252F-
1(2013)
F. Zhang, H. Morkoc, et al. “InGaN based multi-double heterostructure light emitting
diodes with electron injector layers”, Proc. of SPIE, 8625,86251J-1 (2013)
F. Zhang, H. Morkoc, et al. “Impact of ballistic electron transport on efficiency of
InGaN based light emitting diodes”, Proc. SPIE, 793917 (2011)
V.Avrutin, F. Zhang, H. Morkoc, et al. “Saga of efficiency degradation at high injection
in InGaN light emitting diodes”, Turkish J. Phys. 38, 3, 269(2014)
D. Rosales, F. Zhang, H. Morkoc, et al. “ Excitonic recombination dynamics in non-
polar GaN/AlGaN quantum wells”, J. Appl. Phys. 115, 073510 (2014)
V. Avrutin, F. Zhang, H. Morkoc, et al. “InGaN light-emitting diodes: Efficiency-
limiting processes at high injection”, J. Vac. Sci. & Tech., A31, 050809 (2013)
S. Okur, F. Zhang, H. Morkoç, et al. “Microscopic distribution of extended defects and
blockage of threading dislocations by stacking faults in semipolar (1 101) GaN revealed
from spatially resolved luminescence”, Appl. Phys. Lett., 103,2119089 (2013)
N. Izyumskaya, F. Zhang, H. Morkoç, et al. “Optical studies of strain and defect
distribution in semipolar (1 101) GaN on patterned Si substrates”, J. Appl. Phys., 114,
113502 (2013)
S. Okur, F. Zhang, H. Morkoc, et al. “GaN-based vertical cavities with all dielectric
reflectors by epitaxial lateral overgrowth”, Jap. J. Appl. Phys. Vol.52, Issue: 8 (2013)
L. Ardaravicius, F. Zhang, H. Morkoc, et al. “Plasmon-controlled o ptimum gate bias for
GaN heterostructure field-effect transistors”, Semicond. Sci. Technol. 28 (2013)
122
R. A. Ferreyra, F. Zhang, H. Morkoc, et al. “Microwave performance of
AlGaN/AlN/GaN -based single and coupled channels HFETs”, Proc. of SPIE, Vol. 8625
86252B-1 (2013)
C. Zhu, F. Zhang, H. Morkoc, et al. “Investigation of microwave and noise properties of
InAlN/GaN HFETs after electrical stress: role of surface effects ”, Proc. of SPIE Vol.
8625 86252H-1 (2013)
X. Li, F. Zhang, H. Morkoc, et al. “Improved quantum efficiency in InGaN light
emitting diodes with multi-double-heterostructure active regions”, Appl. Phys. Lett., 101,
041115 (2012)
AB. Yankovich, F. Zhang, H. Morkoc, et al. “Hexagonal-based pyramid void defects in
GaN and InGaN”, J. Appl. Phys., 111, 023517 (2012)
C. Zhu, F. Zhang, H. Morkoc, et al. “Degradation and phase noise of InAlN/AlN/GaN
heterojunction field effect transistors: Implications for hot electron/phonon effects”, Appl.
Phys. Lett., 101, 103502 (2012)
X. Li, F. Zhang, H. Morkoc, et al. “Impact of active layer design on InGaN radiative
recombination coefficient and LED performance”, J. Appl. Phys., 111, 063112 (2012)
L. Ardaravicius, F. Zhang, H. Morkoc, et al. “Hot-electron drift velocity in
AlGaN/AlN/AlGaN/GaN camelback Channel”, Semicond. Sci. Technol. 27 (2012)
C. Zhu, F. Zhang, H. Morkoc, et al. “Reduction of Flicker Noise in AlGaN/GaN-Based
HFETs after High Electric-Field Stress”, Elec. Dev. Lett., IEEE 32 (11), 1513, (2011)
X. Li, F. Zhang, H. Morkoc, et al. “On the quantum efficiency of InGaN light emitting
diodes: Effects of active layer design, electron cooler, and electron blocking layer”,
Physic status solidi (a). Vol.208, Issue: 12 (2011)
123
X. Li, F. Zhang, H. Morkoc, et al. “On the reduction of efficiency loss in polar c plane
and nonpolar m plane InGaN light emitting diodes”, Physic status solidi (c). Vol. 8, Issue
5, (2010)
124
Recommended